dheraingoud commited on
Commit
2e7b8ca
·
1 Parent(s): f7872b8

sync: upstream 8dfff832 (v6.18.5) + custom addons (KeyRotator, 401/403 rotation, sol/terra/luna, ScannerTrap, health, Py3.12 compat)

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. pyproject.toml +19 -15
  2. src/free_claude_code/api/admin_routes.py +61 -1
  3. src/free_claude_code/api/admin_static/admin.css +10 -5
  4. src/free_claude_code/api/admin_static/admin.js +167 -36
  5. src/free_claude_code/api/admin_static/code_sessions.js +86 -43
  6. src/free_claude_code/api/admin_static/index.html +27 -0
  7. src/free_claude_code/api/admin_static/session_ui.js +10 -1
  8. src/free_claude_code/api/app.py +35 -7
  9. src/free_claude_code/api/command_utils.py +0 -165
  10. src/free_claude_code/api/dependencies.py +29 -0
  11. src/free_claude_code/api/detection.py +1 -141
  12. src/free_claude_code/api/generation_paths.py +16 -0
  13. src/free_claude_code/api/handlers/gemini.py +123 -0
  14. src/free_claude_code/api/handlers/messages.py +11 -56
  15. src/free_claude_code/api/optimization_handlers.py +0 -163
  16. src/free_claude_code/api/ports.py +6 -0
  17. src/free_claude_code/api/request_errors.py +7 -1
  18. src/free_claude_code/api/request_ids.py +3 -1
  19. src/free_claude_code/api/request_lifetime.py +7 -2
  20. src/free_claude_code/api/request_outcomes.py +14 -6
  21. src/free_claude_code/api/response_streams.py +38 -1
  22. src/free_claude_code/api/routes.py +78 -2
  23. src/free_claude_code/api/stream_delivery.py +18 -12
  24. src/free_claude_code/api/stream_frames.py +1 -11
  25. src/free_claude_code/api/tool_call_buffer.py +4 -1
  26. src/free_claude_code/application/code_sessions/service.py +2 -2
  27. src/free_claude_code/application/execution.py +62 -3
  28. src/free_claude_code/application/model_metadata.py +2 -1
  29. src/free_claude_code/application/ports.py +17 -1
  30. src/free_claude_code/application/reasoning.py +15 -3
  31. src/free_claude_code/application/routing.py +33 -1
  32. src/free_claude_code/cli/desktop_entrypoint.py +10 -0
  33. src/free_claude_code/cli/doctor.py +21 -2
  34. src/free_claude_code/cli/launchers/agy.py +95 -0
  35. src/free_claude_code/cli/launchers/opencode_config.py +3 -0
  36. src/free_claude_code/cli/macos_app.py +148 -0
  37. src/free_claude_code/cli/uvicorn_server.py +28 -0
  38. src/free_claude_code/config/admin/manifest.py +0 -40
  39. src/free_claude_code/config/admin/provider_manifest.py +4 -2
  40. src/free_claude_code/config/constants.py +1 -1
  41. src/free_claude_code/config/env_migrations.py +1 -3
  42. src/free_claude_code/config/loader.py +2 -4
  43. src/free_claude_code/config/model_refs.py +38 -6
  44. src/free_claude_code/config/nim.py +0 -122
  45. src/free_claude_code/config/settings.py +1 -28
  46. src/free_claude_code/core/anthropic/tokens.py +6 -3
  47. src/free_claude_code/core/async_rwlock.py +3 -3
  48. src/free_claude_code/core/continuation_overlap.py +32 -19
  49. src/free_claude_code/core/continuation_stream.py +17 -0
  50. src/free_claude_code/core/delivered_response.py +61 -16
pyproject.toml CHANGED
@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
4
 
5
  [project]
6
  name = "free-claude-code"
7
- version = "6.16.4"
8
  description = "Local proxy connecting coding agents to OpenAI-compatible AI providers"
9
  readme = "README.md"
10
  license = "AGPL-3.0-only"
@@ -17,28 +17,28 @@ dependencies = [
17
  "httpx[socks]>=0.28.1",
18
  "httpx2[socks]>=2.7.0,<3",
19
  "markdown-it-py>=4.2.0",
20
- "pydantic>=2.13.5",
21
  "python-dotenv>=1.2.4",
22
  "tiktoken>=0.13.0",
23
  "python-telegram-bot>=22.8",
24
  "discord.py>=2.7.1",
25
- "openai>=3.25.0",
26
  "loguru>=0.7.0",
27
- "aiohttp>=3.14.4",
28
  "jsonschema>=4.25.0",
29
- "json5>=0.15.0",
30
- "tomlkit>=0.13.3",
31
  "google-auth[requests]>=2.56.3",
32
  "requests[socks]>=2.34.2",
33
  "pystray>=0.19.5; sys_platform == 'win32' or sys_platform == 'darwin'",
34
- "Pillow>=12.0.0; sys_platform == 'win32' or sys_platform == 'darwin'",
35
- "github-copilot-sdk>=1.0.16",
36
  "simplejson>=4.2.0",
37
- "grpcio>=1.84.0",
38
  "grpcio-tools>=1.81.1",
39
  "nvidia-riva-client>=2.26.0",
40
  "pyperclip>=1.11.0",
41
- "sqlalchemy>=2.1.3",
42
  "ruamel-yaml>=0.19.1",
43
  ]
44
 
@@ -55,6 +55,7 @@ fcc-hermes = "free_claude_code.cli.launchers.hermes:launch"
55
  fcc-dsh = "free_claude_code.cli.launchers.dsh:launch"
56
  fcc-grok = "free_claude_code.cli.launchers.grok:launch"
57
  fcc-muse = "free_claude_code.cli.launchers.muse:launch"
 
58
  fcc-aider = "free_claude_code.cli.launchers.aider:launch"
59
 
60
  [project.gui-scripts]
@@ -63,11 +64,14 @@ fcc-desktop = "free_claude_code.cli.desktop_entrypoint:launch"
63
  [project.optional-dependencies]
64
  voice_local = [
65
  "torch>=2.14.1",
66
- "transformers>=5.19.0",
67
  "accelerate>=1.15.0",
68
  "librosa>=1.0.0",
69
  ]
70
 
 
 
 
71
  [tool.hatch.build.targets.wheel]
72
  packages = ["src/free_claude_code"]
73
 
@@ -76,7 +80,7 @@ packages = ["src/free_claude_code"]
76
  "scripts/update/fcc-update.cmd" = "fcc-update.cmd"
77
 
78
  [tool.uv]
79
- required-version = ">=0.12.13"
80
  cache-keys = [{ file = "pyproject.toml" }, { git = { commit = true, tags = true } }]
81
 
82
  [tool.uv.sources]
@@ -92,10 +96,10 @@ dev = [
92
  "pytest>=9.1.1",
93
  "pytest-asyncio>=1.4.0",
94
  "pytest-cov>=7.1.0",
95
- "ty>=0.0.84",
96
- "ruff>=0.16.10",
97
  "pytest-xdist>=3.8.0",
98
- "pytest-playwright>=0.9.0",
99
  "pyyaml>=6.0.3",
100
  "pytest-split>=0.11.0",
101
  ]
 
4
 
5
  [project]
6
  name = "free-claude-code"
7
+ version = "6.18.5"
8
  description = "Local proxy connecting coding agents to OpenAI-compatible AI providers"
9
  readme = "README.md"
10
  license = "AGPL-3.0-only"
 
17
  "httpx[socks]>=0.28.1",
18
  "httpx2[socks]>=2.7.0,<3",
19
  "markdown-it-py>=4.2.0",
20
+ "pydantic>=2.14.1",
21
  "python-dotenv>=1.2.4",
22
  "tiktoken>=0.13.0",
23
  "python-telegram-bot>=22.8",
24
  "discord.py>=2.7.1",
25
+ "openai>=3.37.0",
26
  "loguru>=0.7.0",
27
+ "aiohttp>=3.14.5",
28
  "jsonschema>=4.25.0",
29
+ "json5>=0.17.3",
30
+ "tomlkit>=0.15.2",
31
  "google-auth[requests]>=2.56.3",
32
  "requests[socks]>=2.34.2",
33
  "pystray>=0.19.5; sys_platform == 'win32' or sys_platform == 'darwin'",
34
+ "Pillow>=13.0.0; sys_platform == 'win32' or sys_platform == 'darwin'",
35
+ "github-copilot-sdk>=1.0.26",
36
  "simplejson>=4.2.0",
37
+ "grpcio>=1.84.1",
38
  "grpcio-tools>=1.81.1",
39
  "nvidia-riva-client>=2.26.0",
40
  "pyperclip>=1.11.0",
41
+ "sqlalchemy>=2.1.6",
42
  "ruamel-yaml>=0.19.1",
43
  ]
44
 
 
55
  fcc-dsh = "free_claude_code.cli.launchers.dsh:launch"
56
  fcc-grok = "free_claude_code.cli.launchers.grok:launch"
57
  fcc-muse = "free_claude_code.cli.launchers.muse:launch"
58
+ fcc-agy = "free_claude_code.cli.launchers.agy:launch"
59
  fcc-aider = "free_claude_code.cli.launchers.aider:launch"
60
 
61
  [project.gui-scripts]
 
64
  [project.optional-dependencies]
65
  voice_local = [
66
  "torch>=2.14.1",
67
+ "transformers>=5.20.2",
68
  "accelerate>=1.15.0",
69
  "librosa>=1.0.0",
70
  ]
71
 
72
+ [tool.hatch.build]
73
+ artifacts = ["src/free_claude_code/assets/macos/fcc-desktop-*"]
74
+
75
  [tool.hatch.build.targets.wheel]
76
  packages = ["src/free_claude_code"]
77
 
 
80
  "scripts/update/fcc-update.cmd" = "fcc-update.cmd"
81
 
82
  [tool.uv]
83
+ required-version = ">=0.13.6"
84
  cache-keys = [{ file = "pyproject.toml" }, { git = { commit = true, tags = true } }]
85
 
86
  [tool.uv.sources]
 
96
  "pytest>=9.1.1",
97
  "pytest-asyncio>=1.4.0",
98
  "pytest-cov>=7.1.0",
99
+ "ty>=0.0.91",
100
+ "ruff>=0.17.2",
101
  "pytest-xdist>=3.8.0",
102
+ "pytest-playwright>=0.10.0",
103
  "pyyaml>=6.0.3",
104
  "pytest-split>=0.11.0",
105
  ]
src/free_claude_code/api/admin_routes.py CHANGED
@@ -3,6 +3,7 @@ from __future__ import annotations
3
 
4
  from collections.abc import Awaitable, Callable, Mapping
5
  from pathlib import Path
 
6
 
7
  import httpx
8
  from fastapi import (
@@ -300,6 +301,33 @@ async def vscode_chat_status(
300
  return await _integration_response(services.admin.vscode_chat_status)
301
 
302
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
303
  @router.post("/admin/api/integrations/vscode-chat/connect")
304
  async def connect_vscode_chat(
305
  request: Request,
@@ -551,9 +579,13 @@ async def _check_local_provider(
551
  "status": "missing_url",
552
  "label": "Missing URL",
553
  "base_url": base_url,
 
 
 
554
  }
555
 
556
  url = f"{clean_url}{path}"
 
557
  try:
558
  async with httpx.AsyncClient(timeout=1.5) as client:
559
  response = await client.get(url)
@@ -564,6 +596,9 @@ async def _check_local_provider(
564
  "label": "Reachable" if ok else "Offline",
565
  "base_url": base_url,
566
  "status_code": response.status_code,
 
 
 
567
  }
568
  except Exception as exc:
569
  logger.debug(
@@ -571,15 +606,40 @@ async def _check_local_provider(
571
  provider_id,
572
  type(exc).__name__,
573
  )
 
574
  return {
575
  "provider_id": provider_id,
576
  "status": "offline",
577
  "label": "Offline",
578
  "base_url": base_url,
579
- "message": _LOCAL_PROVIDER_CHECK_FAILURE_MESSAGE,
 
 
 
580
  }
581
 
582
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
583
  def _require_connected_account_provider(provider_id: str) -> None:
584
  descriptor = PROVIDER_CATALOG.get(provider_id)
585
  if (
 
3
 
4
  from collections.abc import Awaitable, Callable, Mapping
5
  from pathlib import Path
6
+ from time import perf_counter
7
 
8
  import httpx
9
  from fastapi import (
 
301
  return await _integration_response(services.admin.vscode_chat_status)
302
 
303
 
304
+ @router.get("/admin/api/integrations/antigravity")
305
+ async def antigravity_status(
306
+ request: Request,
307
+ services: ApiServices = Depends(get_services),
308
+ ):
309
+ require_loopback_admin(request)
310
+ return await _integration_response(services.admin.antigravity_status)
311
+
312
+
313
+ @router.post("/admin/api/integrations/antigravity/connect")
314
+ async def connect_antigravity(
315
+ request: Request,
316
+ services: ApiServices = Depends(get_services),
317
+ ):
318
+ require_loopback_admin(request)
319
+ return await _integration_response(services.admin.connect_antigravity)
320
+
321
+
322
+ @router.post("/admin/api/integrations/antigravity/disconnect")
323
+ async def disconnect_antigravity(
324
+ request: Request,
325
+ services: ApiServices = Depends(get_services),
326
+ ):
327
+ require_loopback_admin(request)
328
+ return await _integration_response(services.admin.disconnect_antigravity)
329
+
330
+
331
  @router.post("/admin/api/integrations/vscode-chat/connect")
332
  async def connect_vscode_chat(
333
  request: Request,
 
579
  "status": "missing_url",
580
  "label": "Missing URL",
581
  "base_url": base_url,
582
+ "probe_path": path,
583
+ "latency_ms": None,
584
+ "error_type": None,
585
  }
586
 
587
  url = f"{clean_url}{path}"
588
+ started = perf_counter()
589
  try:
590
  async with httpx.AsyncClient(timeout=1.5) as client:
591
  response = await client.get(url)
 
596
  "label": "Reachable" if ok else "Offline",
597
  "base_url": base_url,
598
  "status_code": response.status_code,
599
+ "probe_path": path,
600
+ "latency_ms": round((perf_counter() - started) * 1000, 2),
601
+ "error_type": None if ok else "http_error",
602
  }
603
  except Exception as exc:
604
  logger.debug(
 
606
  provider_id,
607
  type(exc).__name__,
608
  )
609
+ error_type, message = _local_provider_failure(exc)
610
  return {
611
  "provider_id": provider_id,
612
  "status": "offline",
613
  "label": "Offline",
614
  "base_url": base_url,
615
+ "message": message,
616
+ "probe_path": path,
617
+ "latency_ms": round((perf_counter() - started) * 1000, 2),
618
+ "error_type": error_type,
619
  }
620
 
621
 
622
+ def _local_provider_failure(exc: Exception) -> tuple[str, str]:
623
+ if isinstance(exc, httpx.TimeoutException):
624
+ return (
625
+ "timeout",
626
+ "The local provider availability check timed out. Check that its server is responsive and retry.",
627
+ )
628
+ if isinstance(exc, (httpx.ConnectError, httpx.ProxyError)):
629
+ return "connection_error", _LOCAL_PROVIDER_CHECK_FAILURE_MESSAGE
630
+ if isinstance(exc, (httpx.InvalidURL, httpx.UnsupportedProtocol)):
631
+ return (
632
+ "invalid_url",
633
+ "Check the local provider's HTTP or HTTPS base URL and port.",
634
+ )
635
+ if isinstance(exc, httpx.RequestError):
636
+ return (
637
+ "transport_error",
638
+ "The connection ended before the availability check completed. Retry and check the local server logs.",
639
+ )
640
+ return "unexpected_error", _LOCAL_PROVIDER_CHECK_FAILURE_MESSAGE
641
+
642
+
643
  def _require_connected_account_provider(provider_id: str) -> None:
644
  descriptor = PROVIDER_CATALOG.get(provider_id)
645
  if (
src/free_claude_code/api/admin_static/admin.css CHANGED
@@ -361,7 +361,8 @@ button.startup-busy::before,
361
  .provider-dialog .field-grid { grid-template-columns: minmax(0, 1fr); }
362
  .provider-dialog .field.advanced-field { display: flex; }
363
  .provider-dialog .provider-actions:not(:empty),
364
- .provider-dialog .provider-check-result:not([hidden]) { margin-bottom: 20px; }
 
365
  .provider-dialog .message-area:empty { display: none; }
366
  .provider-dialog-footer { display: flex; justify-content: flex-end; gap: 12px; margin-top: 24px; }
367
 
@@ -411,24 +412,28 @@ button.startup-busy::before,
411
  overflow-wrap: anywhere;
412
  }
413
 
414
- .provider-check-result {
 
415
  color: var(--muted);
416
  font-size: 12px;
417
  line-height: 1.4;
418
  overflow-wrap: anywhere;
419
  }
420
 
421
- .provider-check-result.ok {
 
422
  color: var(--ok);
423
  }
424
 
425
- .provider-check-result.error {
 
426
  color: var(--error);
427
  }
428
 
429
  .provider-meta.error { color: var(--error); }
430
 
431
- .provider-check-result.checking {
 
432
  color: var(--info);
433
  }
434
 
 
361
  .provider-dialog .field-grid { grid-template-columns: minmax(0, 1fr); }
362
  .provider-dialog .field.advanced-field { display: flex; }
363
  .provider-dialog .provider-actions:not(:empty),
364
+ .provider-dialog .provider-check-result:not([hidden]),
365
+ .provider-dialog .provider-availability-result:not([hidden]) { margin-bottom: 20px; }
366
  .provider-dialog .message-area:empty { display: none; }
367
  .provider-dialog-footer { display: flex; justify-content: flex-end; gap: 12px; margin-top: 24px; }
368
 
 
412
  overflow-wrap: anywhere;
413
  }
414
 
415
+ .provider-check-result,
416
+ .provider-availability-result {
417
  color: var(--muted);
418
  font-size: 12px;
419
  line-height: 1.4;
420
  overflow-wrap: anywhere;
421
  }
422
 
423
+ .provider-check-result.ok,
424
+ .provider-availability-result.ok {
425
  color: var(--ok);
426
  }
427
 
428
+ .provider-check-result.error,
429
+ .provider-availability-result.error {
430
  color: var(--error);
431
  }
432
 
433
  .provider-meta.error { color: var(--error); }
434
 
435
+ .provider-check-result.checking,
436
+ .provider-availability-result.checking {
437
  color: var(--info);
438
  }
439
 
src/free_claude_code/api/admin_static/admin.js CHANGED
@@ -19,7 +19,7 @@ const state = {
19
  startupRequest: null,
20
  startupTimer: null,
21
  startupAgain: false,
22
- codeCatalogRetry: false,
23
  modelOptionsRequest: 0,
24
  activeView: viewFromLocation(),
25
  };
@@ -142,6 +142,7 @@ function renderStartup() {
142
  renderProviderCheckResult(provider.provider_id);
143
  });
144
  renderClaudeIntegration();
 
145
  renderVSCodeChatIntegration();
146
  renderJetBrainsIntegration();
147
  renderCodexIntegration();
@@ -177,8 +178,8 @@ async function refreshStartup() {
177
  const changed = !previous || previous.instance_id !== result.instance_id ||
178
  JSON.stringify(previous.startup) !== JSON.stringify(current);
179
  if (changed) void hydrateModelOptions();
180
- if (changed || state.codeCatalogRetry) state.codeCatalogRetry = await window.CodeSessions?.refresh(result) === false;
181
- pending = state.codeCatalogRetry || current.catalog === "starting" || current.catalog_file === "starting" ||
182
  current.code?.state === "starting" || current.messaging?.state === "starting" ||
183
  Object.values(current.integrations || {}).some((update) => update.state === "starting") ||
184
  Object.values(current.providers || {}).includes("starting");
@@ -209,11 +210,13 @@ window.addEventListener("pagehide", () => {
209
  });
210
 
211
  async function load({ providersOnly = false } = {}) {
 
212
  state.localStatusRequest = null;
213
  state.startupRequest?.controller.abort();
214
  state.startupRequest = null;
215
  showMessage("Loading admin config");
216
  const config = await api("/admin/api/config");
 
217
  config.custom_providers ||= [];
218
  config.provider_status.push(...config.custom_providers.map((provider) => ({ ...provider, kind: "custom", status: "configured", settings_keys: [], missing_configuration_keys: [] })));
219
  state.config = config;
@@ -229,11 +232,10 @@ async function load({ providersOnly = false } = {}) {
229
  byId("configPath").textContent = config.paths.managed;
230
  void refreshLocalStatus(config);
231
  void refreshStartup();
232
- await Promise.all([
233
- refreshConnectedAccounts(),
234
- hydrateModelOptions(),
235
- providersOnly ? window.CodeSessions.refresh() : window.CodeSessions.initialize(api),
236
- ]);
237
  if (state.config !== config) return;
238
  updateDirtyState();
239
  showMessage("");
@@ -265,7 +267,7 @@ function setActiveView(viewId, { scroll = false } = {}) {
265
  state.activeView = activeView.id;
266
  byId("pageTitle").textContent = activeView.title;
267
  renderStartup();
268
- if (activeView.id === "code") state.codeCatalogRetry = true;
269
  void refreshStartup();
270
  const sessionActive = activeView.id === "code";
271
  document.querySelector(".app-shell").classList.toggle("session-active", sessionActive);
@@ -295,6 +297,7 @@ function setActiveView(viewId, { scroll = false } = {}) {
295
  if (activeView.id === "code") window.CodeSessions.activate(window.location.pathname);
296
  else window.CodeSessions.deactivate();
297
  if (activeView.id === "integrations") {
 
298
  refreshClaudeIntegration();
299
  refreshClaudeOnboarding();
300
  refreshVSCodeChatIntegration();
@@ -407,6 +410,10 @@ function updateProviderCard(provider) {
407
  result.className = "provider-check-result";
408
  result.dataset.providerCheckResult = provider.provider_id;
409
  result.hidden = true;
 
 
 
 
410
  const actions = document.createElement("div");
411
  actions.className = "provider-actions";
412
  if (oauth) populateConnectedAccountActions(provider, status, actions);
@@ -418,7 +425,7 @@ function updateProviderCard(provider) {
418
  settings.setAttribute("aria-controls", "providerDialog");
419
  actions.appendChild(settings);
420
  }
421
- card.replaceChildren(title, meta, result, actions);
422
  const next = [...grid.children].find((other) => other !== card &&
423
  connectedAccountName(provider).localeCompare(providerDisplayName(other.dataset.provider), "en", { sensitivity: "base" }) < 0);
424
  if (card.parentElement !== grid || card.nextElementSibling !== (next || null)) grid.insertBefore(card, next || null);
@@ -489,6 +496,7 @@ async function openCustomProviderDialog(providerId = null) {
489
  byId("providerDialogTitle").textContent = providerId ? provider.display_name : "Add provider";
490
  byId("providerMessage").textContent = "";
491
  byId("providerDialogCheck").hidden = true;
 
492
  const fields = byId("providerFields");
493
  fields.replaceChildren();
494
  const formats = (state.config.custom_reasoning_formats || {})[provider.api_format] || ["provider_default"];
@@ -553,10 +561,8 @@ function renderProviderDialogActions(provider) {
553
  actions.appendChild(button);
554
  }
555
  const check = oauth ? null : providerCheckResult(provider.provider_id);
556
- const result = byId("providerDialogCheck");
557
- result.className = `provider-check-result ${check?.status || ""}`;
558
- result.textContent = check?.message || "";
559
- result.hidden = !check?.message;
560
  byId("saveProvider").hidden = !provider.settings_keys?.length;
561
  renderStartup();
562
  }
@@ -779,14 +785,22 @@ function providerCheckResult(providerId) {
779
  // An explicit check takes precedence over background discovery and reachability.
780
  if (check?.source === "manual") return check;
781
  const discovery = state.startup?.startup?.providers?.[providerId];
782
- if (discovery === "starting") return { status: "checking", message: "Checking models…" };
783
- if (discovery === "failed") {
784
- return { status: "error", message: "Could not load models. Check the provider's settings and retry." };
785
- }
786
- if (discovery === "ready") {
787
- return { status: "ok", message: modelCountMessage(state.startup.cached_models[providerId]?.length || 0) };
 
 
 
 
 
 
 
788
  }
789
- return check;
 
790
  }
791
 
792
  function updateProviderCheckResult(providerId, status, message, source = "manual") {
@@ -794,8 +808,17 @@ function updateProviderCheckResult(providerId, status, message, source = "manual
794
  renderProviderCheckResult(providerId);
795
  }
796
 
 
 
 
 
 
 
 
 
797
  function renderProviderCheckResult(providerId) {
798
- const { status = "", message = "" } = providerCheckResult(providerId) || {};
 
799
  const card = document.querySelector(`[data-provider="${providerId}"]`);
800
  if (!card) return;
801
  const provider = connectedAccountDescriptor(providerId);
@@ -810,15 +833,11 @@ function renderProviderCheckResult(providerId) {
810
  if (account?.state === "error") meta.classList.add("error");
811
  return;
812
  }
813
- const result = card.querySelector(".provider-check-result");
814
- result.className = `provider-check-result ${status}`;
815
- result.textContent = message;
816
- result.hidden = !message;
817
  if (state.providerId === providerId) {
818
- const modalResult = byId("providerDialogCheck");
819
- modalResult.className = result.className;
820
- modalResult.textContent = message;
821
- modalResult.hidden = !message;
822
  }
823
  }
824
 
@@ -1544,7 +1563,7 @@ async function showRestartNotice() {
1544
  async function restoreSavedProvider(providerId, checks) {
1545
  if (!connectedAccountDescriptor(providerId)) return;
1546
  await openProviderDialog(providerId);
1547
- showCredentialErrors(checks);
1548
  byId("providerMessage").textContent = "Saved. Some keys could not be verified.";
1549
  byId("providerMessage").className = "message-area warn";
1550
  }
@@ -1598,7 +1617,7 @@ async function apply(providerId = null, customAction = null) {
1598
  applied = true;
1599
  const warningChecks = checks.filter((check) => check.status === "unverified");
1600
  const warnings = warningChecks.map((check) =>
1601
- `${state.fields.get(check.key)?.label || "API keys"}, Key ${check.index + 1}: ${check.message}`
1602
  );
1603
  const savedProviderId = warnings.length && providerId && mutation?.action !== "delete" ? result.custom_provider_id || providerId : null;
1604
  if (providerId && !savedProviderId) byId("providerDialog").close();
@@ -1653,8 +1672,14 @@ async function refreshLocalStatus(config) {
1653
  const provider = await api(`/admin/api/providers/${providerId}/local-status`);
1654
  if (state.localStatusRequest !== request || !request.providerIds.has(providerId)) return;
1655
  if (provider.status === "missing_url") return;
 
 
 
 
 
 
1656
  if (provider.status === "reachable") {
1657
- updateProviderCheckResult(providerId, "ok", `Reachable: ${provider.base_url}`, "availability");
1658
  return;
1659
  }
1660
  const detail = provider.message
@@ -1662,7 +1687,7 @@ async function refreshLocalStatus(config) {
1662
  : provider.status_code
1663
  ? `${provider.base_url} returned HTTP ${provider.status_code}`
1664
  : "The local provider did not respond.";
1665
- updateProviderCheckResult(providerId, "error", `Unavailable: ${detail}`, "availability");
1666
  } catch {
1667
  if (state.localStatusRequest !== request || !request.providerIds.has(providerId)) return;
1668
  updateProviderCheckResult(providerId, "error", "Availability check failed. Use Test to retry.", "availability");
@@ -1693,7 +1718,7 @@ async function testProvider(providerId, button) {
1693
  "ok",
1694
  modelCountMessage(result.models.length),
1695
  );
1696
- await hydrateModelOptions();
1697
  } else {
1698
  updateProviderCheckResult(
1699
  providerId,
@@ -1733,7 +1758,7 @@ async function loadModelOptions(refresh = false) {
1733
  method: refresh ? "POST" : "GET",
1734
  });
1735
  if (request === state.modelOptionsRequest && config === state.config) { state.modelLabels = new Map(Object.entries(result.model_labels || {})); setModelOptions(result.models); }
1736
- if (refresh && window.CodeSessions) await window.CodeSessions.refresh();
1737
  return result;
1738
  }
1739
 
@@ -2077,6 +2102,112 @@ claudeIntegrationDialog.addEventListener("click", (event) => {
2077
  }
2078
  });
2079
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2080
  const vscodeChatIntegrationDialog = byId("vscodeChatIntegrationDialog");
2081
  const vscodeChatIntegration = { connected: null, busy: false, paths: null, update: null };
2082
  const vscodeChatIntegrationPath = "/admin/api/integrations/vscode-chat";
 
19
  startupRequest: null,
20
  startupTimer: null,
21
  startupAgain: false,
22
+ configRequest: 0,
23
  modelOptionsRequest: 0,
24
  activeView: viewFromLocation(),
25
  };
 
142
  renderProviderCheckResult(provider.provider_id);
143
  });
144
  renderClaudeIntegration();
145
+ renderAntigravityIntegration();
146
  renderVSCodeChatIntegration();
147
  renderJetBrainsIntegration();
148
  renderCodexIntegration();
 
178
  const changed = !previous || previous.instance_id !== result.instance_id ||
179
  JSON.stringify(previous.startup) !== JSON.stringify(current);
180
  if (changed) void hydrateModelOptions();
181
+ if (changed) void window.CodeSessions?.refresh(result);
182
+ pending = current.catalog === "starting" || current.catalog_file === "starting" ||
183
  current.code?.state === "starting" || current.messaging?.state === "starting" ||
184
  Object.values(current.integrations || {}).some((update) => update.state === "starting") ||
185
  Object.values(current.providers || {}).includes("starting");
 
210
  });
211
 
212
  async function load({ providersOnly = false } = {}) {
213
+ const request = ++state.configRequest;
214
  state.localStatusRequest = null;
215
  state.startupRequest?.controller.abort();
216
  state.startupRequest = null;
217
  showMessage("Loading admin config");
218
  const config = await api("/admin/api/config");
219
+ if (request !== state.configRequest) return;
220
  config.custom_providers ||= [];
221
  config.provider_status.push(...config.custom_providers.map((provider) => ({ ...provider, kind: "custom", status: "configured", settings_keys: [], missing_configuration_keys: [] })));
222
  state.config = config;
 
232
  byId("configPath").textContent = config.paths.managed;
233
  void refreshLocalStatus(config);
234
  void refreshStartup();
235
+ void refreshConnectedAccounts();
236
+ void hydrateModelOptions();
237
+ if (providersOnly) void window.CodeSessions.refresh();
238
+ else window.CodeSessions.initialize(api);
 
239
  if (state.config !== config) return;
240
  updateDirtyState();
241
  showMessage("");
 
267
  state.activeView = activeView.id;
268
  byId("pageTitle").textContent = activeView.title;
269
  renderStartup();
270
+ if (activeView.id === "code") void window.CodeSessions.refresh();
271
  void refreshStartup();
272
  const sessionActive = activeView.id === "code";
273
  document.querySelector(".app-shell").classList.toggle("session-active", sessionActive);
 
297
  if (activeView.id === "code") window.CodeSessions.activate(window.location.pathname);
298
  else window.CodeSessions.deactivate();
299
  if (activeView.id === "integrations") {
300
+ refreshAntigravityIntegration();
301
  refreshClaudeIntegration();
302
  refreshClaudeOnboarding();
303
  refreshVSCodeChatIntegration();
 
410
  result.className = "provider-check-result";
411
  result.dataset.providerCheckResult = provider.provider_id;
412
  result.hidden = true;
413
+ const availability = document.createElement("span");
414
+ availability.className = "provider-availability-result";
415
+ availability.dataset.providerAvailabilityResult = provider.provider_id;
416
+ availability.hidden = true;
417
  const actions = document.createElement("div");
418
  actions.className = "provider-actions";
419
  if (oauth) populateConnectedAccountActions(provider, status, actions);
 
425
  settings.setAttribute("aria-controls", "providerDialog");
426
  actions.appendChild(settings);
427
  }
428
+ card.replaceChildren(title, meta, result, availability, actions);
429
  const next = [...grid.children].find((other) => other !== card &&
430
  connectedAccountName(provider).localeCompare(providerDisplayName(other.dataset.provider), "en", { sensitivity: "base" }) < 0);
431
  if (card.parentElement !== grid || card.nextElementSibling !== (next || null)) grid.insertBefore(card, next || null);
 
496
  byId("providerDialogTitle").textContent = providerId ? provider.display_name : "Add provider";
497
  byId("providerMessage").textContent = "";
498
  byId("providerDialogCheck").hidden = true;
499
+ renderProviderCheckMessage(byId("providerDialogAvailability"), null);
500
  const fields = byId("providerFields");
501
  fields.replaceChildren();
502
  const formats = (state.config.custom_reasoning_formats || {})[provider.api_format] || ["provider_default"];
 
561
  actions.appendChild(button);
562
  }
563
  const check = oauth ? null : providerCheckResult(provider.provider_id);
564
+ renderProviderCheckMessage(byId("providerDialogCheck"), check);
565
+ renderProviderCheckMessage(byId("providerDialogAvailability"), check?.availability);
 
 
566
  byId("saveProvider").hidden = !provider.settings_keys?.length;
567
  renderStartup();
568
  }
 
785
  // An explicit check takes precedence over background discovery and reachability.
786
  if (check?.source === "manual") return check;
787
  const discovery = state.startup?.startup?.providers?.[providerId];
788
+ let result;
789
+ if (discovery === "starting") {
790
+ result = { status: "checking", message: "Checking models…" };
791
+ } else if (discovery === "failed") {
792
+ result = { status: "error", message: state.startup.startup.provider_errors?.[providerId] || "Could not load models. Check the provider's settings and retry." };
793
+ } else if (discovery === "ready") {
794
+ result = { status: "ok", message: modelCountMessage(state.startup.cached_models[providerId]?.length || 0) };
795
+ } else if (discovery === "unavailable") {
796
+ return {
797
+ status: "",
798
+ message: "Local server unavailable",
799
+ availability: check ? { ...check, status: "" } : undefined,
800
+ };
801
  }
802
+ // Model discovery and endpoint reachability are separate observations.
803
+ return result ? { ...result, availability: check } : check;
804
  }
805
 
806
  function updateProviderCheckResult(providerId, status, message, source = "manual") {
 
808
  renderProviderCheckResult(providerId);
809
  }
810
 
811
+ function renderProviderCheckMessage(element, check) {
812
+ for (const status of ["ok", "error", "checking"]) {
813
+ element.classList.toggle(status, check?.status === status);
814
+ }
815
+ element.textContent = check?.message || "";
816
+ element.hidden = !check?.message;
817
+ }
818
+
819
  function renderProviderCheckResult(providerId) {
820
+ const check = providerCheckResult(providerId);
821
+ const { status = "" } = check || {};
822
  const card = document.querySelector(`[data-provider="${providerId}"]`);
823
  if (!card) return;
824
  const provider = connectedAccountDescriptor(providerId);
 
833
  if (account?.state === "error") meta.classList.add("error");
834
  return;
835
  }
836
+ renderProviderCheckMessage(card.querySelector(".provider-check-result"), check);
837
+ renderProviderCheckMessage(card.querySelector(".provider-availability-result"), check?.availability);
 
 
838
  if (state.providerId === providerId) {
839
+ renderProviderCheckMessage(byId("providerDialogCheck"), check);
840
+ renderProviderCheckMessage(byId("providerDialogAvailability"), check?.availability);
 
 
841
  }
842
  }
843
 
 
1563
  async function restoreSavedProvider(providerId, checks) {
1564
  if (!connectedAccountDescriptor(providerId)) return;
1565
  await openProviderDialog(providerId);
1566
+ showCredentialErrors(checks.map((check) => ({ ...check, message: `Key saved. ${check.message}` })));
1567
  byId("providerMessage").textContent = "Saved. Some keys could not be verified.";
1568
  byId("providerMessage").className = "message-area warn";
1569
  }
 
1617
  applied = true;
1618
  const warningChecks = checks.filter((check) => check.status === "unverified");
1619
  const warnings = warningChecks.map((check) =>
1620
+ `${state.fields.get(check.key)?.label || "API keys"}, Key ${check.index + 1}: Key saved. ${check.message}`
1621
  );
1622
  const savedProviderId = warnings.length && providerId && mutation?.action !== "delete" ? result.custom_provider_id || providerId : null;
1623
  if (providerId && !savedProviderId) byId("providerDialog").close();
 
1672
  const provider = await api(`/admin/api/providers/${providerId}/local-status`);
1673
  if (state.localStatusRequest !== request || !request.providerIds.has(providerId)) return;
1674
  if (provider.status === "missing_url") return;
1675
+ const diagnostics = [];
1676
+ if (provider.probe_path) diagnostics.push(`GET ${provider.probe_path}`);
1677
+ if (typeof provider.latency_ms === "number" && Number.isFinite(provider.latency_ms)) {
1678
+ diagnostics.push(`${provider.latency_ms} ms`);
1679
+ }
1680
+ const suffix = diagnostics.length ? ` (${diagnostics.join(", ")})` : "";
1681
  if (provider.status === "reachable") {
1682
+ updateProviderCheckResult(providerId, "ok", `Reachable: ${provider.base_url}${suffix}`, "availability");
1683
  return;
1684
  }
1685
  const detail = provider.message
 
1687
  : provider.status_code
1688
  ? `${provider.base_url} returned HTTP ${provider.status_code}`
1689
  : "The local provider did not respond.";
1690
+ updateProviderCheckResult(providerId, "error", `Unavailable: ${detail}${suffix}`, "availability");
1691
  } catch {
1692
  if (state.localStatusRequest !== request || !request.providerIds.has(providerId)) return;
1693
  updateProviderCheckResult(providerId, "error", "Availability check failed. Use Test to retry.", "availability");
 
1718
  "ok",
1719
  modelCountMessage(result.models.length),
1720
  );
1721
+ void hydrateModelOptions();
1722
  } else {
1723
  updateProviderCheckResult(
1724
  providerId,
 
1758
  method: refresh ? "POST" : "GET",
1759
  });
1760
  if (request === state.modelOptionsRequest && config === state.config) { state.modelLabels = new Map(Object.entries(result.model_labels || {})); setModelOptions(result.models); }
1761
+ if (refresh && window.CodeSessions) void window.CodeSessions.refresh();
1762
  return result;
1763
  }
1764
 
 
2102
  }
2103
  });
2104
 
2105
+ const antigravityIntegrationDialog = byId("antigravityIntegrationDialog");
2106
+ const antigravityIntegration = { connected: null, canDisconnect: false, disconnecting: false, busy: false, paths: null };
2107
+ const antigravityIntegrationPath = "/admin/api/integrations/antigravity";
2108
+
2109
+ function renderAntigravityIntegration() {
2110
+ const { connected, canDisconnect, disconnecting, busy, paths } = antigravityIntegration;
2111
+ const action = connected ? "Disconnect" : "Connect";
2112
+ byId("openAntigravityIntegration").textContent = busy ? "Loading…" : connected === null ? "Retry" : action;
2113
+ byId("openAntigravityIntegration").disabled = busy;
2114
+ byId("openAntigravityIntegration").setAttribute("aria-busy", String(busy));
2115
+ byId("openAntigravityIntegration").className = connected && !busy ? "danger-button" : "primary-button";
2116
+ byId("disconnectAntigravityIntegration").hidden = connected !== false || !canDisconnect;
2117
+ byId("disconnectAntigravityIntegration").disabled = busy;
2118
+ byId("confirmAntigravityIntegration").textContent = busy ? "Saving…" : disconnecting ? "Disconnect" : "Connect";
2119
+ byId("confirmAntigravityIntegration").disabled = busy || connected === null;
2120
+ byId("confirmAntigravityIntegration").className = disconnecting ? "danger-button" : "primary-button";
2121
+ byId("antigravityIntegrationDescription").textContent = disconnecting
2122
+ ? connected
2123
+ ? "Restore Antigravity CLI's previous provider and remove FCC's models. Running sessions can continue. Connect again before your next fcc-agy launch."
2124
+ : "Remove FCC's saved connection and models. Keep your current Antigravity CLI provider."
2125
+ : "Switch Antigravity CLI to the Gemini API provider and add FCC's models. Then launch with fcc-agy.";
2126
+ const files = byId("antigravityIntegrationFiles");
2127
+ files.replaceChildren();
2128
+ for (const path of Object.values(paths || {})) {
2129
+ const item = document.createElement("li");
2130
+ const code = document.createElement("code");
2131
+ code.textContent = path;
2132
+ item.appendChild(code);
2133
+ files.appendChild(item);
2134
+ }
2135
+ }
2136
+
2137
+ async function refreshAntigravityIntegration() {
2138
+ const request = beginIntegrationCheck(antigravityIntegration);
2139
+ if (!request) return;
2140
+ integrationMessage("antigravityIntegrationMessage", "");
2141
+ renderAntigravityIntegration();
2142
+ try {
2143
+ const result = await api(antigravityIntegrationPath);
2144
+ if (!currentIntegrationCheck(antigravityIntegration, request)) return;
2145
+ antigravityIntegration.connected = result.connected;
2146
+ antigravityIntegration.canDisconnect = result.can_disconnect;
2147
+ antigravityIntegration.paths = result.paths;
2148
+ } catch (error) {
2149
+ if (!currentIntegrationCheck(antigravityIntegration, request)) return;
2150
+ antigravityIntegration.connected = null;
2151
+ integrationMessage("antigravityIntegrationMessage", error.message, true);
2152
+ } finally {
2153
+ finishIntegrationCheck(antigravityIntegration, request, renderAntigravityIntegration);
2154
+ }
2155
+ }
2156
+
2157
+ function openAntigravityIntegration(disconnect) {
2158
+ antigravityIntegration.disconnecting = disconnect;
2159
+ renderAntigravityIntegration();
2160
+ integrationMessage("antigravityIntegrationDialogMessage", "");
2161
+ antigravityIntegrationDialog.showModal();
2162
+ }
2163
+
2164
+ byId("openAntigravityIntegration").addEventListener("click", () => {
2165
+ if (antigravityIntegration.connected === null) {
2166
+ void refreshAntigravityIntegration();
2167
+ return;
2168
+ }
2169
+ openAntigravityIntegration(antigravityIntegration.connected);
2170
+ });
2171
+ byId("disconnectAntigravityIntegration").addEventListener("click", () => openAntigravityIntegration(true));
2172
+ byId("confirmAntigravityIntegration").addEventListener("click", async () => {
2173
+ if (byId("confirmAntigravityIntegration").disabled) return;
2174
+ const disconnect = antigravityIntegration.disconnecting;
2175
+ const config = state.config;
2176
+ let failed = false;
2177
+ antigravityIntegration.busy = true;
2178
+ renderAntigravityIntegration();
2179
+ integrationMessage("antigravityIntegrationDialogMessage", "");
2180
+ integrationMessage("antigravityIntegrationMessage", "");
2181
+ try {
2182
+ const result = await api(`${antigravityIntegrationPath}/${disconnect ? "disconnect" : "connect"}`, { method: "POST" });
2183
+ if (config !== state.config) return;
2184
+ antigravityIntegration.connected = result.connected;
2185
+ antigravityIntegration.canDisconnect = result.can_disconnect;
2186
+ antigravityIntegration.paths = result.paths;
2187
+ antigravityIntegrationDialog.close();
2188
+ integrationMessage("antigravityIntegrationMessage", disconnect
2189
+ ? "Disconnected. FCC's connection removed."
2190
+ : "Connected. Launch Antigravity CLI with fcc-agy.");
2191
+ } catch (error) {
2192
+ if (config !== state.config) return;
2193
+ failed = true;
2194
+ integrationMessage("antigravityIntegrationDialogMessage", error.message, true);
2195
+ integrationMessage("antigravityIntegrationMessage", error.message, true);
2196
+ } finally {
2197
+ antigravityIntegration.busy = false;
2198
+ renderAntigravityIntegration();
2199
+ if (failed || config !== state.config) void refreshAntigravityIntegration();
2200
+ }
2201
+ });
2202
+ byId("closeAntigravityIntegration").addEventListener("click", () => antigravityIntegrationDialog.close());
2203
+ antigravityIntegrationDialog.addEventListener("click", (event) => {
2204
+ if (event.target !== antigravityIntegrationDialog) return;
2205
+ const bounds = antigravityIntegrationDialog.getBoundingClientRect();
2206
+ if (event.clientX < bounds.left || event.clientX > bounds.right || event.clientY < bounds.top || event.clientY > bounds.bottom) {
2207
+ antigravityIntegrationDialog.close();
2208
+ }
2209
+ });
2210
+
2211
  const vscodeChatIntegrationDialog = byId("vscodeChatIntegrationDialog");
2212
  const vscodeChatIntegration = { connected: null, busy: false, paths: null, update: null };
2213
  const vscodeChatIntegrationPath = "/admin/api/integrations/vscode-chat";
src/free_claude_code/api/admin_static/code_sessions.js CHANGED
@@ -14,9 +14,15 @@
14
  catalogLoaded = false,
15
  catalogPhase = "starting",
16
  catalogVersion = null,
 
 
17
  catalogInstance = null,
18
  catalogRequest = 0,
19
  settingsPending = false;
 
 
 
 
20
  let visibleIds = [],
21
  query = "",
22
  libraryToken = 0,
@@ -364,32 +370,60 @@
364
  record.nextBefore = data.next_before;
365
  render(!before && outputChanged);
366
  }
367
- function olderCatalog(candidate) {
368
- return candidate && catalogVersion && (candidate.generation_id < catalogVersion.generation_id ||
369
- (candidate.generation_id === catalogVersion.generation_id && candidate.catalog_revision < catalogVersion.catalog_revision));
370
  }
371
- async function bootstrap() {
372
  const token = syncToken;
373
  const request = ++catalogRequest;
 
 
374
  try {
375
- const data = await api(`${base}/bootstrap`);
376
- if (request !== catalogRequest || token !== syncToken || (epoch && data.epoch !== epoch) || olderCatalog(data.startup)) return false;
377
  if (data.startup) catalogVersion = data.startup;
378
  available = data.available;
379
  catalogPhase = data.startup?.catalog || "ready";
380
- if (catalogPhase === "ready" || !catalogLoaded) {
381
- catalog = data.models;
382
- harnesses = data.harnesses;
383
- }
384
- if (catalogPhase === "ready") catalogLoaded = true;
385
  availabilityNotice = data.message || "";
386
  return true;
387
  } catch (error) {
388
- if (request === catalogRequest && token === syncToken) {
389
  availabilityNotice = error.message;
390
  if (!catalogLoaded) catalogPhase = "failed";
391
  }
392
- throw error;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
393
  }
394
  }
395
 
@@ -402,6 +436,7 @@
402
  if (epoch !== readyData.epoch) {
403
  if (epoch) {
404
  catalogVersion = null;
 
405
  catalogLoaded = false;
406
  catalogPhase = "starting";
407
  }
@@ -429,7 +464,7 @@
429
  for (const summary of readyData.sessions)
430
  merge({ ...summary, cursor: readyData.cursor });
431
  render();
432
- void bootstrap().then(() => render()).catch(() => render());
433
  try {
434
  const results = await Promise.allSettled([
435
  list(),
@@ -479,17 +514,12 @@
479
  source.addEventListener(type, (event) => {
480
  if (feed === source) receive(type, event);
481
  });
482
- const disconnected = async () => {
483
  if (feed !== source) return;
484
  restart();
485
  // A stopped store cannot open its event feed. Read its availability
486
  // so that the retry screen can explain why it is unavailable.
487
- try {
488
- await bootstrap();
489
- render();
490
- } catch {
491
- /* The scheduled connection retries the reads. */
492
- }
493
  };
494
  source.addEventListener("feed.resync_required", disconnected);
495
  source.onerror = disconnected;
@@ -591,11 +621,7 @@
591
  if (selected === id) notice = error.message;
592
  }
593
  if (error.status === 400 || error.status === 409) {
594
- try {
595
- await bootstrap();
596
- } catch (failure) {
597
- restart(failure.message);
598
- }
599
  }
600
  } finally {
601
  sends.delete(id);
@@ -837,7 +863,9 @@
837
  if (providerDraft) return "Choose a model for the selected provider.";
838
  if (!catalogLoaded) return catalogPhase === "starting" ? "Models are loading…" : "Model list unavailable.";
839
  const model = catalog.find((model) => model.id === record.session.model);
840
- if (!model) return "Selected model is unavailable. Choose another model.";
 
 
841
  if (
842
  record.session.reasoning_effort &&
843
  !model.reasoning_efforts.includes(record.session.reasoning_effort)
@@ -846,6 +874,12 @@
846
  return "";
847
  }
848
 
 
 
 
 
 
 
849
  function remove() {
850
  const record = records.get(selected),
851
  id = selected,
@@ -1241,8 +1275,8 @@
1241
  stop.hidden = !isBusy;
1242
  stop.disabled = !connected || record?.run?.stop_requested;
1243
  send.textContent = saved(selected).pending ? "Retry Send" : "Send";
1244
- send.classList.toggle("startup-busy", !catalogLoaded && catalogPhase === "starting");
1245
- send.setAttribute("aria-busy", String(!catalogLoaded && catalogPhase === "starting"));
1246
  send.disabled =
1247
  !ready(record) ||
1248
  !available ||
@@ -1268,12 +1302,12 @@
1268
  record.session?.status === "deleting" ||
1269
  isBusy ||
1270
  pending(record);
1271
- const disabled =
1272
  !ready(record) ||
1273
  isBusy ||
1274
  pending(record) ||
1275
- settingsPending ||
1276
- !catalogLoaded;
1277
  const provider =
1278
  providerDraft?.provider || record?.session?.provider_id || "";
1279
  if (modelControl.group.dataset.provider !== provider) {
@@ -1298,7 +1332,7 @@
1298
  );
1299
  harnessControl.select.disabled = true;
1300
  modeControl.select.value = record?.session?.mode || "config";
1301
- modeControl.select.disabled = disabled;
1302
  const model = catalog.find((model) => model.id === record?.session?.model),
1303
  effort =
1304
  record?.session?.reasoning_effort ||
@@ -1661,6 +1695,18 @@
1661
  dismissDialog();
1662
  }
1663
  window.addEventListener("pagehide", deactivate);
 
 
 
 
 
 
 
 
 
 
 
 
1664
  window.CodeSessions = {
1665
  setProviderNames(providers) {
1666
  providerNames = new Map(providers.map((provider) => [provider.provider_id, provider.display_name]));
@@ -1675,18 +1721,15 @@
1675
  async refresh(status) {
1676
  if (!api) return false;
1677
  if (status?.startup) {
1678
- if (catalogInstance !== status.instance_id) catalogVersion = null;
 
 
 
 
1679
  catalogInstance = status.instance_id;
1680
- if (!olderCatalog(status.startup)) catalogVersion = status.startup;
1681
- }
1682
- try {
1683
- const refreshed = await bootstrap();
1684
- render();
1685
- return refreshed;
1686
- } catch {
1687
- render();
1688
- return false;
1689
  }
 
1690
  },
1691
  };
1692
  document.addEventListener("pointerdown", (event) => {
 
14
  catalogLoaded = false,
15
  catalogPhase = "starting",
16
  catalogVersion = null,
17
+ catalogTarget = null,
18
+ catalogProviders = {},
19
  catalogInstance = null,
20
  catalogRequest = 0,
21
  settingsPending = false;
22
+ let catalogFlight = null,
23
+ catalogAgain = false,
24
+ catalogRetryTimer = null,
25
+ catalogController = null;
26
  let visibleIds = [],
27
  query = "",
28
  libraryToken = 0,
 
370
  record.nextBefore = data.next_before;
371
  render(!before && outputChanged);
372
  }
373
+ function olderCatalog(candidate, version = catalogTarget) {
374
+ return candidate && version && (candidate.generation_id < version.generation_id ||
375
+ (candidate.generation_id === version.generation_id && candidate.catalog_revision < version.catalog_revision));
376
  }
377
+ async function readBootstrap() {
378
  const token = syncToken;
379
  const request = ++catalogRequest;
380
+ const controller = new AbortController();
381
+ catalogController = controller;
382
  try {
383
+ const data = await api(`${base}/bootstrap`, { signal: controller.signal });
384
+ if (request !== catalogRequest || token !== syncToken || (epoch && data.epoch !== epoch) || olderCatalog(data.startup) || olderCatalog(data.startup, catalogVersion)) return false;
385
  if (data.startup) catalogVersion = data.startup;
386
  available = data.available;
387
  catalogPhase = data.startup?.catalog || "ready";
388
+ catalogProviders = data.startup?.providers || {};
389
+ catalog = data.models;
390
+ harnesses = data.harnesses;
391
+ catalogLoaded = true;
 
392
  availabilityNotice = data.message || "";
393
  return true;
394
  } catch (error) {
395
+ if (error.name !== "AbortError" && request === catalogRequest && token === syncToken) {
396
  availabilityNotice = error.message;
397
  if (!catalogLoaded) catalogPhase = "failed";
398
  }
399
+ return false;
400
+ } finally {
401
+ if (catalogController === controller) catalogController = null;
402
+ }
403
+ }
404
+
405
+ async function bootstrap() {
406
+ if (!api || document.hidden) return false;
407
+ catalogAgain = true;
408
+ clearTimeout(catalogRetryTimer);
409
+ catalogRetryTimer = null;
410
+ if (catalogFlight) return catalogFlight;
411
+ catalogFlight = (async () => {
412
+ while (catalogAgain && !document.hidden) {
413
+ catalogAgain = false;
414
+ const refreshed = await readBootstrap();
415
+ render();
416
+ if (!refreshed && !catalogAgain) {
417
+ if (!document.hidden) catalogRetryTimer = setTimeout(() => void bootstrap(), 500);
418
+ return false;
419
+ }
420
+ }
421
+ return true;
422
+ })();
423
+ try {
424
+ return await catalogFlight;
425
+ } finally {
426
+ catalogFlight = null;
427
  }
428
  }
429
 
 
436
  if (epoch !== readyData.epoch) {
437
  if (epoch) {
438
  catalogVersion = null;
439
+ catalogTarget = null;
440
  catalogLoaded = false;
441
  catalogPhase = "starting";
442
  }
 
464
  for (const summary of readyData.sessions)
465
  merge({ ...summary, cursor: readyData.cursor });
466
  render();
467
+ void bootstrap();
468
  try {
469
  const results = await Promise.allSettled([
470
  list(),
 
514
  source.addEventListener(type, (event) => {
515
  if (feed === source) receive(type, event);
516
  });
517
+ const disconnected = () => {
518
  if (feed !== source) return;
519
  restart();
520
  // A stopped store cannot open its event feed. Read its availability
521
  // so that the retry screen can explain why it is unavailable.
522
+ void bootstrap();
 
 
 
 
 
523
  };
524
  source.addEventListener("feed.resync_required", disconnected);
525
  source.onerror = disconnected;
 
621
  if (selected === id) notice = error.message;
622
  }
623
  if (error.status === 400 || error.status === 409) {
624
+ await bootstrap();
 
 
 
 
625
  }
626
  } finally {
627
  sends.delete(id);
 
863
  if (providerDraft) return "Choose a model for the selected provider.";
864
  if (!catalogLoaded) return catalogPhase === "starting" ? "Models are loading…" : "Model list unavailable.";
865
  const model = catalog.find((model) => model.id === record.session.model);
866
+ if (!model) return catalogProviders[record.session.provider_id] === "starting"
867
+ ? "Models for the selected provider are loading…"
868
+ : "Selected model is unavailable. Choose another model.";
869
  if (
870
  record.session.reasoning_effort &&
871
  !model.reasoning_efforts.includes(record.session.reasoning_effort)
 
874
  return "";
875
  }
876
 
877
+ function selectionLoading(record) {
878
+ if (!catalogLoaded) return catalogPhase === "starting";
879
+ return !!record?.session && !catalog.some((model) => model.id === record.session.model) &&
880
+ catalogProviders[record.session.provider_id] === "starting";
881
+ }
882
+
883
  function remove() {
884
  const record = records.get(selected),
885
  id = selected,
 
1275
  stop.hidden = !isBusy;
1276
  stop.disabled = !connected || record?.run?.stop_requested;
1277
  send.textContent = saved(selected).pending ? "Retry Send" : "Send";
1278
+ send.classList.toggle("startup-busy", selectionLoading(record));
1279
+ send.setAttribute("aria-busy", String(selectionLoading(record)));
1280
  send.disabled =
1281
  !ready(record) ||
1282
  !available ||
 
1302
  record.session?.status === "deleting" ||
1303
  isBusy ||
1304
  pending(record);
1305
+ const sessionDisabled =
1306
  !ready(record) ||
1307
  isBusy ||
1308
  pending(record) ||
1309
+ settingsPending;
1310
+ const disabled = sessionDisabled || !catalogLoaded || !catalog.length;
1311
  const provider =
1312
  providerDraft?.provider || record?.session?.provider_id || "";
1313
  if (modelControl.group.dataset.provider !== provider) {
 
1332
  );
1333
  harnessControl.select.disabled = true;
1334
  modeControl.select.value = record?.session?.mode || "config";
1335
+ modeControl.select.disabled = sessionDisabled;
1336
  const model = catalog.find((model) => model.id === record?.session?.model),
1337
  effort =
1338
  record?.session?.reasoning_effort ||
 
1695
  dismissDialog();
1696
  }
1697
  window.addEventListener("pagehide", deactivate);
1698
+ function pauseCatalog() {
1699
+ clearTimeout(catalogRetryTimer);
1700
+ catalogRetryTimer = null;
1701
+ catalogAgain = false;
1702
+ ++catalogRequest;
1703
+ catalogController?.abort();
1704
+ }
1705
+ window.addEventListener("pagehide", pauseCatalog);
1706
+ document.addEventListener("visibilitychange", () => {
1707
+ if (document.hidden) pauseCatalog();
1708
+ else void bootstrap();
1709
+ });
1710
  window.CodeSessions = {
1711
  setProviderNames(providers) {
1712
  providerNames = new Map(providers.map((provider) => [provider.provider_id, provider.display_name]));
 
1721
  async refresh(status) {
1722
  if (!api) return false;
1723
  if (status?.startup) {
1724
+ if (catalogInstance !== status.instance_id) {
1725
+ catalogTarget = null;
1726
+ catalogVersion = null;
1727
+ ++catalogRequest;
1728
+ }
1729
  catalogInstance = status.instance_id;
1730
+ if (!olderCatalog(status.startup)) catalogTarget = status.startup;
 
 
 
 
 
 
 
 
1731
  }
1732
+ return bootstrap();
1733
  },
1734
  };
1735
  document.addEventListener("pointerdown", (event) => {
src/free_claude_code/api/admin_static/index.html CHANGED
@@ -65,6 +65,32 @@
65
  </section>
66
 
67
  <section id="view-integrations" class="admin-view provider-grid" data-view="integrations" hidden>
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
68
  <article class="provider-strip integration-card">
69
  <div class="section-heading">
70
  <div>
@@ -259,6 +285,7 @@
259
  <div id="providerDialogBody">
260
  <div id="providerDialogActions" class="provider-actions"></div>
261
  <div id="providerDialogCheck" class="provider-check-result" role="status" hidden></div>
 
262
  <div id="providerFields" class="field-grid"></div>
263
  </div>
264
  <div id="providerMessage" class="message-area" role="status"></div>
 
65
  </section>
66
 
67
  <section id="view-integrations" class="admin-view provider-grid" data-view="integrations" hidden>
68
+ <article class="provider-strip integration-card">
69
+ <div class="section-heading">
70
+ <div>
71
+ <h3>Antigravity CLI</h3>
72
+ <p>Connect Antigravity CLI to use FCC's models with fcc-agy.</p>
73
+ </div>
74
+ </div>
75
+ <p id="antigravityIntegrationMessage" class="message-area" role="status" hidden></p>
76
+ <div class="integration-actions">
77
+ <button id="openAntigravityIntegration" class="primary-button" type="button" aria-busy="true" disabled>Loading…</button>
78
+ <button id="disconnectAntigravityIntegration" class="danger-button" type="button" hidden>Disconnect</button>
79
+ </div>
80
+ </article>
81
+ <dialog id="antigravityIntegrationDialog" class="integration-dialog" aria-labelledby="antigravityIntegrationTitle" aria-describedby="antigravityIntegrationDescription">
82
+ <div class="section-heading">
83
+ <h3 id="antigravityIntegrationTitle">Antigravity CLI</h3>
84
+ <button id="closeAntigravityIntegration" class="ghost-button" type="button" aria-label="Close" autofocus>×</button>
85
+ </div>
86
+ <p id="antigravityIntegrationDescription">Switch Antigravity CLI to the Gemini API provider and add FCC's models. Then launch with fcc-agy.</p>
87
+ <div class="integration-files">
88
+ <strong>Files to modify</strong>
89
+ <ul id="antigravityIntegrationFiles"></ul>
90
+ </div>
91
+ <button id="confirmAntigravityIntegration" class="primary-button" type="button">Connect</button>
92
+ <p id="antigravityIntegrationDialogMessage" class="message-area error" role="alert" hidden></p>
93
+ </dialog>
94
  <article class="provider-strip integration-card">
95
  <div class="section-heading">
96
  <div>
 
285
  <div id="providerDialogBody">
286
  <div id="providerDialogActions" class="provider-actions"></div>
287
  <div id="providerDialogCheck" class="provider-check-result" role="status" hidden></div>
288
+ <div id="providerDialogAvailability" class="provider-availability-result" role="status" hidden></div>
289
  <div id="providerFields" class="field-grid"></div>
290
  </div>
291
  <div id="providerMessage" class="message-area" role="status"></div>
src/free_claude_code/api/admin_static/session_ui.js CHANGED
@@ -87,7 +87,8 @@
87
  input.autocomplete = "off";
88
  input.value = value;
89
  input.setAttribute("aria-label", "Selected model");
90
- let committed = value;
 
91
  const combobox = new window.FccModelCombobox(input, {
92
  listboxId: `${id}-options`,
93
  label: "model",
@@ -114,6 +115,14 @@
114
  input.disabled = disabled;
115
  combobox.toggle.disabled = disabled;
116
  if (disabled) combobox.close();
 
 
 
 
 
 
 
 
117
  },
118
  };
119
  }
 
87
  input.autocomplete = "off";
88
  input.value = value;
89
  input.setAttribute("aria-label", "Selected model");
90
+ let committed = value,
91
+ optionsSignature = JSON.stringify(values());
92
  const combobox = new window.FccModelCombobox(input, {
93
  listboxId: `${id}-options`,
94
  label: "model",
 
115
  input.disabled = disabled;
116
  combobox.toggle.disabled = disabled;
117
  if (disabled) combobox.close();
118
+ const signature = JSON.stringify(values());
119
+ if (combobox.isOpen && signature !== optionsSignature) {
120
+ const active = combobox.visibleOptions[combobox.activeIndex]?.dataset.value;
121
+ combobox.render(combobox.query);
122
+ const index = combobox.visibleOptions.findIndex((option) => option.dataset.value === active);
123
+ if (index >= 0) combobox.setActive(index, false);
124
+ }
125
+ optionsSignature = signature;
126
  },
127
  };
128
  }
src/free_claude_code/api/app.py CHANGED
@@ -2,11 +2,15 @@ from __future__ import annotations
2
  """Pure FastAPI application factory."""
3
 
4
  from fastapi import FastAPI, Request
5
- from fastapi.exception_handlers import request_validation_exception_handler
 
 
 
6
  from fastapi.exceptions import RequestValidationError
7
  from fastapi.responses import JSONResponse, Response
8
- from starlette.middleware.base import BaseHTTPMiddleware
9
  from loguru import logger
 
 
10
 
11
  from free_claude_code.application.code_sessions import (
12
  CodeConflictError,
@@ -21,6 +25,7 @@ from free_claude_code.core.diagnostics import (
21
  redacted_exception_traceback,
22
  safe_exception_message,
23
  )
 
24
  from free_claude_code.core.openai_responses import openai_error_payload
25
  from free_claude_code.core.trace import (
26
  extract_claude_session_id_from_headers,
@@ -31,6 +36,7 @@ from free_claude_code.core.version import package_version
31
  from .admin_cache import AdminNoStoreMiddleware, attach_admin_no_store
32
  from .admin_routes import router as admin_router
33
  from .code_sessions_routes import router as code_router
 
34
  from .ports import ApiServices
35
  from .request_errors import ordinary_application_error_response
36
  from .request_ids import (
@@ -44,6 +50,7 @@ from .routes import router
44
  from .validation_log import summarize_request_validation_body
45
 
46
 
 
47
  class ScannerTrapMiddleware(BaseHTTPMiddleware):
48
  """Intercept common automated scanner probes and return 200 OK."""
49
 
@@ -68,6 +75,16 @@ def create_app(services: ApiServices) -> FastAPI:
68
  app.include_router(code_router)
69
  app.include_router(router)
70
 
 
 
 
 
 
 
 
 
 
 
71
  @app.exception_handler(CodeError)
72
  async def code_error_handler(request: Request, exc: CodeError):
73
  if isinstance(exc, CodeNotFoundError):
@@ -102,12 +119,23 @@ def create_app(services: ApiServices) -> FastAPI:
102
  event="server.request.validation_failed",
103
  source="api",
104
  path=request.url.path,
105
- query=dict(request.query_params),
 
 
 
106
  error_locs=[list(error.get("loc", ())) for error in exc.errors()],
107
  error_types=[str(error.get("type", "")) for error in exc.errors()],
108
  message_summary=message_summary,
109
  tool_names=tool_names,
110
  )
 
 
 
 
 
 
 
 
111
  if request.url.path == "/admin/api/config/apply":
112
  return JSONResponse(
113
  status_code=422,
@@ -125,9 +153,7 @@ def create_app(services: ApiServices) -> FastAPI:
125
  """Serialize defensive application failures in the selected wire protocol."""
126
  return ordinary_application_error_response(
127
  exc,
128
- wire_api=(
129
- "responses" if request.url.path == "/v1/responses" else "messages"
130
- ),
131
  request_id=get_request_id(request),
132
  )
133
 
@@ -154,7 +180,9 @@ def create_app(services: ApiServices) -> FastAPI:
154
  type(exc).__name__,
155
  )
156
  message = safe_exception_message(exc)
157
- if request.url.path == "/v1/responses":
 
 
158
  content = openai_error_payload(message=message, error_type="api_error")
159
  else:
160
  content = anthropic_error_payload(
 
2
  """Pure FastAPI application factory."""
3
 
4
  from fastapi import FastAPI, Request
5
+ from fastapi.exception_handlers import (
6
+ http_exception_handler,
7
+ request_validation_exception_handler,
8
+ )
9
  from fastapi.exceptions import RequestValidationError
10
  from fastapi.responses import JSONResponse, Response
 
11
  from loguru import logger
12
+ from starlette.exceptions import HTTPException
13
+ from starlette.middleware.base import BaseHTTPMiddleware
14
 
15
  from free_claude_code.application.code_sessions import (
16
  CodeConflictError,
 
25
  redacted_exception_traceback,
26
  safe_exception_message,
27
  )
28
+ from free_claude_code.core.gemini import gemini_error_payload
29
  from free_claude_code.core.openai_responses import openai_error_payload
30
  from free_claude_code.core.trace import (
31
  extract_claude_session_id_from_headers,
 
36
  from .admin_cache import AdminNoStoreMiddleware, attach_admin_no_store
37
  from .admin_routes import router as admin_router
38
  from .code_sessions_routes import router as code_router
39
+ from .generation_paths import generation_api
40
  from .ports import ApiServices
41
  from .request_errors import ordinary_application_error_response
42
  from .request_ids import (
 
50
  from .validation_log import summarize_request_validation_body
51
 
52
 
53
+
54
  class ScannerTrapMiddleware(BaseHTTPMiddleware):
55
  """Intercept common automated scanner probes and return 200 OK."""
56
 
 
75
  app.include_router(code_router)
76
  app.include_router(router)
77
 
78
+ @app.exception_handler(HTTPException)
79
+ async def public_http_error_handler(request: Request, exc: HTTPException):
80
+ if generation_api(request.url.path) == "gemini":
81
+ return JSONResponse(
82
+ status_code=exc.status_code,
83
+ content=gemini_error_payload(exc.status_code, str(exc.detail)),
84
+ headers=exc.headers,
85
+ )
86
+ return await http_exception_handler(request, exc)
87
+
88
  @app.exception_handler(CodeError)
89
  async def code_error_handler(request: Request, exc: CodeError):
90
  if isinstance(exc, CodeNotFoundError):
 
119
  event="server.request.validation_failed",
120
  source="api",
121
  path=request.url.path,
122
+ query={
123
+ key: "[REDACTED]" if key == "key" else value
124
+ for key, value in request.query_params.items()
125
+ },
126
  error_locs=[list(error.get("loc", ())) for error in exc.errors()],
127
  error_types=[str(error.get("type", "")) for error in exc.errors()],
128
  message_summary=message_summary,
129
  tool_names=tool_names,
130
  )
131
+ if generation_api(request.url.path) == "gemini":
132
+ message = "; ".join(
133
+ f"{'.'.join(map(str, error['loc']))}: {error['msg']}"
134
+ for error in exc.errors()
135
+ )
136
+ return JSONResponse(
137
+ status_code=400, content=gemini_error_payload(400, message)
138
+ )
139
  if request.url.path == "/admin/api/config/apply":
140
  return JSONResponse(
141
  status_code=422,
 
153
  """Serialize defensive application failures in the selected wire protocol."""
154
  return ordinary_application_error_response(
155
  exc,
156
+ wire_api=generation_api(request.url.path) or "messages",
 
 
157
  request_id=get_request_id(request),
158
  )
159
 
 
180
  type(exc).__name__,
181
  )
182
  message = safe_exception_message(exc)
183
+ if generation_api(request.url.path) == "gemini":
184
+ content = gemini_error_payload(500, message)
185
+ elif request.url.path == "/v1/responses":
186
  content = openai_error_payload(message=message, error_type="api_error")
187
  else:
188
  content = anthropic_error_payload(
src/free_claude_code/api/command_utils.py DELETED
@@ -1,165 +0,0 @@
1
- from __future__ import annotations
2
- """Command parsing utilities for API optimizations."""
3
-
4
- import re
5
- import shlex
6
-
7
- _ENV_ASSIGNMENT_RE = re.compile(r"^[A-Za-z_][A-Za-z0-9_]*=.*$")
8
-
9
-
10
- def _is_env_assignment(part: str) -> bool:
11
- """Return True when a token is a shell-style env assignment."""
12
- return bool(_ENV_ASSIGNMENT_RE.match(part))
13
-
14
-
15
- def _strip_env_assignments(parts: list[str]) -> list[str]:
16
- """Return command parts after leading shell-style env assignments."""
17
- cmd_start = 0
18
- for i, part in enumerate(parts):
19
- if _is_env_assignment(part):
20
- cmd_start = i + 1
21
- else:
22
- break
23
- return parts[cmd_start:]
24
-
25
-
26
- def extract_command_prefix(command: str) -> str:
27
- """Extract the command prefix for fast prefix detection.
28
-
29
- Parses a shell command safely, handling environment variables and
30
- command injection attempts. Returns the command prefix suitable
31
- for quick identification.
32
-
33
- Returns:
34
- Command prefix (e.g., "git", "git commit", "npm install")
35
- or "none" if no valid command found
36
- """
37
- if "`" in command or "$(" in command:
38
- return "command_injection_detected"
39
-
40
- try:
41
- parts = shlex.split(command, posix=False)
42
- if not parts:
43
- return "none"
44
-
45
- env_prefix = []
46
- cmd_start = 0
47
- for i, part in enumerate(parts):
48
- if _is_env_assignment(part):
49
- env_prefix.append(part)
50
- cmd_start = i + 1
51
- else:
52
- break
53
-
54
- if cmd_start >= len(parts):
55
- return "none"
56
-
57
- cmd_parts = parts[cmd_start:]
58
- if not cmd_parts:
59
- return "none"
60
-
61
- first_word = cmd_parts[0]
62
- two_word_commands = {
63
- "git",
64
- "npm",
65
- "docker",
66
- "kubectl",
67
- "cargo",
68
- "go",
69
- "pip",
70
- "yarn",
71
- }
72
-
73
- if first_word in two_word_commands and len(cmd_parts) > 1:
74
- second_word = cmd_parts[1]
75
- if not second_word.startswith("-"):
76
- return f"{first_word} {second_word}"
77
- return first_word
78
- return first_word if not env_prefix else " ".join(env_prefix) + " " + first_word
79
-
80
- except ValueError:
81
- parts = command.split()
82
- if not parts:
83
- return "none"
84
- cmd_parts = _strip_env_assignments(parts)
85
- return cmd_parts[0] if cmd_parts else "none"
86
-
87
-
88
- def extract_filepaths_from_command(command: str, output: str) -> str:
89
- """Extract file paths from a command locally without API call.
90
-
91
- Determines if the command reads file contents and extracts paths accordingly.
92
- Commands like ls/dir/find just list files, so return empty.
93
- Commands like cat/head/tail actually read contents, so extract the file path.
94
-
95
- Returns:
96
- Filepath extraction result in <filepaths> format
97
- """
98
- listing_commands = {
99
- "ls",
100
- "dir",
101
- "find",
102
- "tree",
103
- "pwd",
104
- "cd",
105
- "mkdir",
106
- "rmdir",
107
- "rm",
108
- }
109
-
110
- reading_commands = {"cat", "head", "tail", "less", "more", "bat", "type"}
111
-
112
- try:
113
- parts = shlex.split(command, posix=False)
114
- if not parts:
115
- return "<filepaths>\n</filepaths>"
116
-
117
- cmd_parts = _strip_env_assignments(parts)
118
- if not cmd_parts:
119
- return "<filepaths>\n</filepaths>"
120
-
121
- base_cmd = cmd_parts[0].split("/")[-1].split("\\")[-1].lower()
122
-
123
- if base_cmd in listing_commands:
124
- return "<filepaths>\n</filepaths>"
125
-
126
- if base_cmd in reading_commands:
127
- filepaths = []
128
- for part in cmd_parts[1:]:
129
- if part.startswith("-"):
130
- continue
131
- filepaths.append(part)
132
-
133
- if filepaths:
134
- paths_str = "\n".join(filepaths)
135
- return f"<filepaths>\n{paths_str}\n</filepaths>"
136
- return "<filepaths>\n</filepaths>"
137
-
138
- if base_cmd == "grep":
139
- flags_with_args = {"-e", "-f", "-m", "-A", "-B", "-C"}
140
- pattern_provided_via_flag = False
141
- positional = []
142
-
143
- skip_next = False
144
- for part in cmd_parts[1:]:
145
- if skip_next:
146
- skip_next = False
147
- continue
148
- if part.startswith("-"):
149
- if part in flags_with_args:
150
- if part in {"-e", "-f"}:
151
- pattern_provided_via_flag = True
152
- skip_next = True
153
- continue
154
- positional.append(part)
155
-
156
- filepaths = positional if pattern_provided_via_flag else positional[1:]
157
- if filepaths:
158
- paths_str = "\n".join(filepaths)
159
- return f"<filepaths>\n{paths_str}\n</filepaths>"
160
- return "<filepaths>\n</filepaths>"
161
-
162
- return "<filepaths>\n</filepaths>"
163
-
164
- except ValueError:
165
- return "<filepaths>\n</filepaths>"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
src/free_claude_code/api/dependencies.py CHANGED
@@ -120,6 +120,35 @@ def require_anthropic_proxy_auth(
120
  )
121
 
122
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
123
  def _proxy_token_matches(
124
  credential: str,
125
  configured_token: str,
 
120
  )
121
 
122
 
123
+ def require_gemini_proxy_auth(
124
+ request: Request,
125
+ settings: Settings = Depends(get_settings),
126
+ ) -> None:
127
+ keys = request.query_params.getlist("key")
128
+ if len(keys) > 1:
129
+ raise HTTPException(
130
+ status_code=400, detail="Repeated key query credentials are unsupported"
131
+ )
132
+ if not settings.proxy_auth_enabled:
133
+ return
134
+ credential = request.headers.get("authorization")
135
+ bearer = credential is not None
136
+ if credential is None:
137
+ credential = request.headers.get("x-goog-api-key")
138
+ if credential is None:
139
+ credential = keys[0] if keys else None
140
+ if credential is None:
141
+ raise HTTPException(
142
+ status_code=401, detail="Missing proxy authentication token"
143
+ )
144
+ if not _proxy_token_matches(
145
+ credential, settings.proxy_auth_token, require_bearer=bearer
146
+ ):
147
+ raise HTTPException(
148
+ status_code=401, detail="Invalid proxy authentication token"
149
+ )
150
+
151
+
152
  def _proxy_token_matches(
153
  credential: str,
154
  configured_token: str,
src/free_claude_code/api/detection.py CHANGED
@@ -1,29 +1,12 @@
1
  from __future__ import annotations
2
- """Request detection utilities for API optimizations.
3
-
4
- Detects quota checks, title generation, prefix detection, safety classifier,
5
- suggestion mode, and filepath extraction requests to enable targeted handling.
6
- """
7
 
8
  from free_claude_code.core.anthropic import (
9
- Message,
10
  MessagesRequest,
11
  extract_text_from_content,
12
  )
13
 
14
 
15
- def _single_user_turn(request_data: MessagesRequest) -> Message | None:
16
- """Return the sole conversational user turn, ignoring system context."""
17
- user_turn: Message | None = None
18
- for message in request_data.messages:
19
- if message.role == "system":
20
- continue
21
- if message.role != "user" or user_turn is not None:
22
- return None
23
- user_turn = message
24
- return user_turn
25
-
26
-
27
  def _request_system_text(request_data: MessagesRequest) -> str:
28
  """Return top-level and inline system text for request-shape detection."""
29
  parts: list[str] = []
@@ -40,65 +23,6 @@ def _request_system_text(request_data: MessagesRequest) -> str:
40
  return "\n".join(parts)
41
 
42
 
43
- def is_quota_check_request(request_data: MessagesRequest) -> bool:
44
- """Check if this is a quota probe request.
45
-
46
- Quota checks are typically simple requests with max_tokens=1
47
- and a single message containing the word "quota".
48
- """
49
- message = _single_user_turn(request_data)
50
- if request_data.max_tokens != 1 or message is None:
51
- return False
52
- text = extract_text_from_content(message.content)
53
- return "quota" in text.lower()
54
-
55
-
56
- def is_title_generation_request(request_data: MessagesRequest) -> bool:
57
- """Check if this is a conversation title generation request.
58
-
59
- Title generation requests are detected by a system prompt containing
60
- title extraction instructions, no tools, and a single user message.
61
-
62
- Matches Claude Code session title prompts (sentence-case title, JSON
63
- \"title\" field, etc.).
64
- """
65
- if request_data.tools:
66
- return False
67
- system_text = _request_system_text(request_data).lower()
68
- if "title" not in system_text:
69
- return False
70
- return "sentence-case title" in system_text or (
71
- "return json" in system_text
72
- and "field" in system_text
73
- and ("coding session" in system_text or "this session" in system_text)
74
- )
75
-
76
-
77
- def is_prefix_detection_request(request_data: MessagesRequest) -> tuple[bool, str]:
78
- """Check if this is a fast prefix detection request.
79
-
80
- Prefix detection requests contain a policy_spec block and
81
- a Command: section for extracting shell command prefixes.
82
-
83
- Returns:
84
- Tuple of (is_prefix_request, command_string)
85
- """
86
- message = _single_user_turn(request_data)
87
- if message is None:
88
- return False, ""
89
-
90
- content = extract_text_from_content(message.content)
91
-
92
- if "<policy_spec>" in content and "Command:" in content:
93
- try:
94
- cmd_start = content.rfind("Command:") + len("Command:")
95
- return True, content[cmd_start:].strip()
96
- except TypeError:
97
- return False, ""
98
-
99
- return False, ""
100
-
101
-
102
  def detect_safety_classifier_stop_sequence(
103
  request_data: MessagesRequest,
104
  ) -> str | None:
@@ -120,67 +44,3 @@ def detect_safety_classifier_stop_sequence(
120
  if "yes</block>" in system_text or "no</block>" in system_text:
121
  return "</block>"
122
  return None
123
-
124
-
125
- def is_suggestion_mode_request(request_data: MessagesRequest) -> bool:
126
- """Check if this is a suggestion mode request.
127
-
128
- Suggestion mode requests contain "[SUGGESTION MODE:" in the user's message,
129
- used for auto-suggesting what the user might type next.
130
- """
131
- for msg in request_data.messages:
132
- if msg.role == "user":
133
- text = extract_text_from_content(msg.content)
134
- if "[SUGGESTION MODE:" in text:
135
- return True
136
- return False
137
-
138
-
139
- def is_filepath_extraction_request(
140
- request_data: MessagesRequest,
141
- ) -> tuple[bool, str, str]:
142
- """Check if this is a filepath extraction request.
143
-
144
- Filepath extraction requests have a single user message with
145
- "Command:" and "Output:" sections, asking to extract file paths
146
- from command output.
147
-
148
- Returns:
149
- Tuple of (is_filepath_request, command, output)
150
- """
151
- message = _single_user_turn(request_data)
152
- if message is None:
153
- return False, "", ""
154
- if request_data.tools:
155
- return False, "", ""
156
-
157
- content = extract_text_from_content(message.content)
158
-
159
- if "Command:" not in content or "Output:" not in content:
160
- return False, "", ""
161
-
162
- # Match if user content OR system block indicates filepath extraction
163
- user_has_filepaths = (
164
- "filepaths" in content.lower() or "<filepaths>" in content.lower()
165
- )
166
- system_text = _request_system_text(request_data)
167
- system_has_extract = (
168
- "extract any file paths" in system_text.lower()
169
- or "file paths that this command" in system_text.lower()
170
- )
171
- if not user_has_filepaths and not system_has_extract:
172
- return False, "", ""
173
-
174
- cmd_start = content.find("Command:") + len("Command:")
175
- output_marker = content.find("Output:", cmd_start)
176
- if output_marker == -1:
177
- return False, "", ""
178
-
179
- command = content[cmd_start:output_marker].strip()
180
- output = content[output_marker + len("Output:") :].strip()
181
-
182
- for marker in ["<", "\n\n"]:
183
- if marker in output:
184
- output = output.split(marker)[0].strip()
185
-
186
- return True, command, output
 
1
  from __future__ import annotations
2
+ """Detect known Claude auto-mode classifier requests."""
 
 
 
 
3
 
4
  from free_claude_code.core.anthropic import (
 
5
  MessagesRequest,
6
  extract_text_from_content,
7
  )
8
 
9
 
 
 
 
 
 
 
 
 
 
 
 
 
10
  def _request_system_text(request_data: MessagesRequest) -> str:
11
  """Return top-level and inline system text for request-shape detection."""
12
  parts: list[str] = []
 
23
  return "\n".join(parts)
24
 
25
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
26
  def detect_safety_classifier_stop_sequence(
27
  request_data: MessagesRequest,
28
  ) -> str | None:
 
44
  if "yes</block>" in system_text or "no</block>" in system_text:
45
  return "</block>"
46
  return None
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
src/free_claude_code/api/generation_paths.py ADDED
@@ -0,0 +1,16 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from __future__ import annotations
2
+ """Public generation route classification shared by HTTP boundaries."""
3
+
4
+ from typing import Literal
5
+
6
+ type WireApi = Literal["messages", "responses", "gemini"]
7
+
8
+
9
+ def generation_api(path: str) -> WireApi | None:
10
+ if path == "/v1/messages":
11
+ return "messages"
12
+ if path == "/v1/responses":
13
+ return "responses"
14
+ if path.startswith("/v1beta/models/"):
15
+ return "gemini"
16
+ return None
src/free_claude_code/api/handlers/gemini.py ADDED
@@ -0,0 +1,123 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from __future__ import annotations
2
+ """Gemini JSON and SSE execution through the shared provider application."""
3
+
4
+ from collections.abc import Mapping
5
+
6
+ from fastapi.responses import JSONResponse, Response
7
+
8
+ from free_claude_code.api.request_errors import (
9
+ log_unexpected_api_exception,
10
+ ordinary_application_error_response,
11
+ )
12
+ from free_claude_code.api.response_streams import (
13
+ ManagedStreamingResponse,
14
+ gemini_sse_streaming_response,
15
+ terminal_execution_error_response,
16
+ trace_terminal_execution_error,
17
+ )
18
+ from free_claude_code.application.errors import ApplicationError
19
+ from free_claude_code.application.execution import ProviderExecutor
20
+ from free_claude_code.application.ports import ModelInfoLookup, ProviderResolver
21
+ from free_claude_code.application.routing import ModelRouter
22
+ from free_claude_code.config.settings import Settings
23
+ from free_claude_code.core.anthropic.streaming.decoder import AnthropicSSEDecoder
24
+ from free_claude_code.core.diagnostics import safe_exception_message
25
+ from free_claude_code.core.failures import find_execution_failure
26
+ from free_claude_code.core.gemini import (
27
+ GeminiConversionError,
28
+ GeminiRequest,
29
+ gemini_error_payload,
30
+ )
31
+ from free_claude_code.core.gemini.response import GeminiResponseAccumulator
32
+
33
+
34
+ class GeminiHandler:
35
+ def __init__(
36
+ self,
37
+ settings: Settings,
38
+ provider_resolver: ProviderResolver,
39
+ *,
40
+ generation_id: int | None = None,
41
+ request_headers: Mapping[str, str] | None = None,
42
+ model_info_lookup: ModelInfoLookup | None = None,
43
+ ) -> None:
44
+ self._settings = settings
45
+ self._router = ModelRouter(settings)
46
+ self._executor = ProviderExecutor(
47
+ provider_resolver,
48
+ progress_timeout_seconds=settings.provider_progress_timeout,
49
+ generation_id=generation_id,
50
+ request_headers=request_headers,
51
+ model_info_lookup=model_info_lookup,
52
+ log_raw_payloads=settings.log_raw_api_payloads,
53
+ )
54
+
55
+ async def create(
56
+ self, request: GeminiRequest, *, streaming: bool, request_id: str
57
+ ) -> Response:
58
+ try:
59
+ routed = self._router.resolve_gemini_request(request)
60
+ body = self._executor.stream_gemini(
61
+ routed,
62
+ request_id=request_id,
63
+ raw_log_payload=lambda: request.model_dump(
64
+ mode="json", by_alias=True, exclude_none=True
65
+ ),
66
+ )
67
+ response = await gemini_sse_streaming_response(
68
+ body,
69
+ request_id=request_id,
70
+ pre_start_error_response=lambda error: self._error(error, request_id),
71
+ )
72
+ if streaming or not isinstance(response, ManagedStreamingResponse):
73
+ return response
74
+ aggregate = GeminiResponseAccumulator()
75
+ decoder = AnthropicSSEDecoder()
76
+ try:
77
+ async for chunk in response.body_iterator:
78
+ text = (
79
+ chunk
80
+ if isinstance(chunk, str)
81
+ else bytes(chunk).decode("utf-8")
82
+ )
83
+ for event in decoder.feed(text):
84
+ aggregate.feed(event.data)
85
+ for event in decoder.finish():
86
+ aggregate.feed(event.data)
87
+ return JSONResponse(content=aggregate.result())
88
+ finally:
89
+ await response.aclose()
90
+ except (Exception, BaseExceptionGroup) as error:
91
+ return self._error(error, request_id)
92
+
93
+ def _error(self, error: BaseException, request_id: str) -> JSONResponse:
94
+ if isinstance(error, ApplicationError):
95
+ return ordinary_application_error_response(
96
+ error, wire_api="gemini", request_id=request_id
97
+ )
98
+ if isinstance(error, GeminiConversionError):
99
+ return JSONResponse(
100
+ status_code=400, content=gemini_error_payload(400, str(error))
101
+ )
102
+ failure = find_execution_failure(error)
103
+ status = failure.status_code if failure is not None else 500
104
+ if failure is None:
105
+ log_unexpected_api_exception(
106
+ self._settings,
107
+ error,
108
+ context="CREATE_GEMINI_ERROR",
109
+ request_id=request_id,
110
+ )
111
+ trace_terminal_execution_error(
112
+ wire_api="gemini",
113
+ request_id=request_id,
114
+ status_code=status,
115
+ error_type=failure.kind.value if failure else "api_error",
116
+ error=error,
117
+ )
118
+ return terminal_execution_error_response(
119
+ status_code=status,
120
+ content=gemini_error_payload(
121
+ status, safe_exception_message(failure or error)
122
+ ),
123
+ )
src/free_claude_code/api/handlers/messages.py CHANGED
@@ -4,13 +4,11 @@ from __future__ import annotations
4
  import asyncio
5
  import sys
6
  from collections.abc import AsyncIterator, Mapping
7
- from dataclasses import dataclass, replace
8
 
9
  from fastapi.responses import JSONResponse, Response
10
- from loguru import logger
11
 
12
  from free_claude_code.api.detection import detect_safety_classifier_stop_sequence
13
- from free_claude_code.api.optimization_handlers import try_optimizations
14
  from free_claude_code.api.request_errors import (
15
  http_status_for_unexpected_api_exception,
16
  log_unexpected_api_exception,
@@ -54,19 +52,6 @@ from free_claude_code.core.trace import close_stream_input, trace_event
54
  from .classifier_response import classifier_response
55
 
56
 
57
- @dataclass(frozen=True)
58
- class _MessagesStreamResult:
59
- body: AsyncIterator[str]
60
-
61
-
62
- @dataclass(frozen=True)
63
- class _MessagesCompleteResult:
64
- response: object
65
-
66
-
67
- _MessagesResult = _MessagesStreamResult | _MessagesCompleteResult
68
-
69
-
70
  class MessagesHandler:
71
  """Handle Anthropic-compatible Messages requests."""
72
 
@@ -141,25 +126,15 @@ class MessagesHandler:
141
  routed.resolved.primary.provider_id,
142
  routed.resolved.primary.provider_model,
143
  )
144
- tool_body = self._web_tools.try_stream_messages(
145
- routed, request_id=request_id
146
- )
147
- result = (
148
- _MessagesStreamResult(tool_body)
149
- if tool_body is not None
150
- else self._intercept_local_optimization(routed)
151
- )
152
- if result is None:
153
- logger.debug("No optimization matched, routing to provider")
154
- result = _MessagesStreamResult(
155
- self._provider_executor.stream_messages(
156
- routed,
157
- raw_log_payload=routed.request.model_dump,
158
- request_id=request_id,
159
- )
160
  )
161
  return await self._to_public_response(
162
- result,
163
  stream=request_data.stream,
164
  request_id=request_id,
165
  hide_reasoning=routed.reasoning.control is ReasoningControl.PREFER_OFF,
@@ -178,21 +153,19 @@ class MessagesHandler:
178
 
179
  async def _to_public_response(
180
  self,
181
- result: _MessagesResult,
182
  *,
183
  stream: bool,
184
  request_id: str,
185
  hide_reasoning: bool = False,
186
  ) -> object:
187
- if isinstance(result, _MessagesCompleteResult):
188
- return result.response
189
  if not stream:
190
  # Non-streaming clients (e.g. Claude Code utility calls) need a
191
  # complete JSON Message; the internal pipeline is always SSE, so
192
  # serving that raw here breaks the client SDK's response parse.
193
  try:
194
  message, error, _complete = await aggregate_anthropic_sse_to_message(
195
- classifier_response(result.body) if hide_reasoning else result.body
196
  )
197
  except GeneratorExit:
198
  raise
@@ -238,7 +211,7 @@ class MessagesHandler:
238
  )
239
  return JSONResponse(content=message)
240
  return await anthropic_sse_streaming_response(
241
- result.body,
242
  pre_start_error_response=lambda exc: self._pre_start_error_response(
243
  exc, request_id=request_id
244
  ),
@@ -361,24 +334,6 @@ class MessagesHandler:
361
  ),
362
  )
363
 
364
- def _intercept_local_optimization(
365
- self, routed: RoutedMessagesRequest
366
- ) -> _MessagesResult | None:
367
- optimized = try_optimizations(
368
- routed.request,
369
- self._settings,
370
- response_model=routed.resolved.original_model,
371
- )
372
- if optimized is None:
373
- return None
374
- trace_event(
375
- stage="routing",
376
- event="free_claude_code.api.optimization.short_circuit",
377
- source="api",
378
- model=routed.resolved.original_model,
379
- )
380
- return _MessagesCompleteResult(optimized)
381
-
382
 
383
  def _stream_error_fields(error: dict[str, object]) -> tuple[str, str]:
384
  raw_type = error.get("type")
 
4
  import asyncio
5
  import sys
6
  from collections.abc import AsyncIterator, Mapping
7
+ from dataclasses import replace
8
 
9
  from fastapi.responses import JSONResponse, Response
 
10
 
11
  from free_claude_code.api.detection import detect_safety_classifier_stop_sequence
 
12
  from free_claude_code.api.request_errors import (
13
  http_status_for_unexpected_api_exception,
14
  log_unexpected_api_exception,
 
52
  from .classifier_response import classifier_response
53
 
54
 
 
 
 
 
 
 
 
 
 
 
 
 
 
55
  class MessagesHandler:
56
  """Handle Anthropic-compatible Messages requests."""
57
 
 
126
  routed.resolved.primary.provider_id,
127
  routed.resolved.primary.provider_model,
128
  )
129
+ body = self._web_tools.try_stream_messages(routed, request_id=request_id)
130
+ if body is None:
131
+ body = self._provider_executor.stream_messages(
132
+ routed,
133
+ raw_log_payload=routed.request.model_dump,
134
+ request_id=request_id,
 
 
 
 
 
 
 
 
 
 
135
  )
136
  return await self._to_public_response(
137
+ body,
138
  stream=request_data.stream,
139
  request_id=request_id,
140
  hide_reasoning=routed.reasoning.control is ReasoningControl.PREFER_OFF,
 
153
 
154
  async def _to_public_response(
155
  self,
156
+ body: AsyncIterator[str],
157
  *,
158
  stream: bool,
159
  request_id: str,
160
  hide_reasoning: bool = False,
161
  ) -> object:
 
 
162
  if not stream:
163
  # Non-streaming clients (e.g. Claude Code utility calls) need a
164
  # complete JSON Message; the internal pipeline is always SSE, so
165
  # serving that raw here breaks the client SDK's response parse.
166
  try:
167
  message, error, _complete = await aggregate_anthropic_sse_to_message(
168
+ classifier_response(body) if hide_reasoning else body
169
  )
170
  except GeneratorExit:
171
  raise
 
211
  )
212
  return JSONResponse(content=message)
213
  return await anthropic_sse_streaming_response(
214
+ body,
215
  pre_start_error_response=lambda exc: self._pre_start_error_response(
216
  exc, request_id=request_id
217
  ),
 
334
  ),
335
  )
336
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
337
 
338
  def _stream_error_fields(error: dict[str, object]) -> tuple[str, str]:
339
  raw_type = error.get("type")
src/free_claude_code/api/optimization_handlers.py DELETED
@@ -1,163 +0,0 @@
1
- from __future__ import annotations
2
- """Optimization handlers for fast-path API responses.
3
-
4
- Each handler returns a MessagesResponse if the request matches and the
5
- optimization is enabled, otherwise None.
6
- """
7
-
8
- import uuid
9
-
10
- from loguru import logger
11
-
12
- from free_claude_code.config.settings import Settings
13
- from free_claude_code.core.anthropic import (
14
- MessagesRequest,
15
- MessagesResponse,
16
- Usage,
17
- )
18
-
19
- from .command_utils import extract_command_prefix, extract_filepaths_from_command
20
- from .detection import (
21
- is_filepath_extraction_request,
22
- is_prefix_detection_request,
23
- is_quota_check_request,
24
- is_suggestion_mode_request,
25
- is_title_generation_request,
26
- )
27
-
28
-
29
- def _text_response(
30
- request_data: MessagesRequest,
31
- text: str,
32
- *,
33
- input_tokens: int,
34
- output_tokens: int,
35
- ) -> MessagesResponse:
36
- return MessagesResponse(
37
- id=f"msg_{uuid.uuid4()}",
38
- model=request_data.model,
39
- content=[{"type": "text", "text": text}],
40
- stop_reason="end_turn",
41
- usage=Usage(input_tokens=input_tokens, output_tokens=output_tokens),
42
- )
43
-
44
-
45
- def try_prefix_detection(
46
- request_data: MessagesRequest, settings: Settings
47
- ) -> MessagesResponse | None:
48
- """Fast prefix detection - return command prefix without API call."""
49
- if not settings.fast_prefix_detection:
50
- return None
51
-
52
- is_prefix_req, command = is_prefix_detection_request(request_data)
53
- if not is_prefix_req:
54
- return None
55
-
56
- logger.info("Optimization: Fast prefix detection request")
57
- return _text_response(
58
- request_data,
59
- extract_command_prefix(command),
60
- input_tokens=100,
61
- output_tokens=5,
62
- )
63
-
64
-
65
- def try_quota_mock(
66
- request_data: MessagesRequest, settings: Settings
67
- ) -> MessagesResponse | None:
68
- """Mock quota probe requests."""
69
- if not settings.enable_network_probe_mock:
70
- return None
71
- if not is_quota_check_request(request_data):
72
- return None
73
-
74
- logger.info("Optimization: Intercepted and mocked quota probe")
75
- return _text_response(
76
- request_data,
77
- "Quota check passed.",
78
- input_tokens=10,
79
- output_tokens=5,
80
- )
81
-
82
-
83
- def try_title_skip(
84
- request_data: MessagesRequest, settings: Settings
85
- ) -> MessagesResponse | None:
86
- """Skip title generation requests."""
87
- if not settings.enable_title_generation_skip:
88
- return None
89
- if not is_title_generation_request(request_data):
90
- return None
91
-
92
- logger.info("Optimization: Skipped title generation request")
93
- return _text_response(
94
- request_data,
95
- "Conversation",
96
- input_tokens=100,
97
- output_tokens=5,
98
- )
99
-
100
-
101
- def try_suggestion_skip(
102
- request_data: MessagesRequest, settings: Settings
103
- ) -> MessagesResponse | None:
104
- """Skip suggestion mode requests."""
105
- if not settings.enable_suggestion_mode_skip:
106
- return None
107
- if not is_suggestion_mode_request(request_data):
108
- return None
109
-
110
- logger.info("Optimization: Skipped suggestion mode request")
111
- return _text_response(
112
- request_data,
113
- "",
114
- input_tokens=100,
115
- output_tokens=1,
116
- )
117
-
118
-
119
- def try_filepath_mock(
120
- request_data: MessagesRequest, settings: Settings
121
- ) -> MessagesResponse | None:
122
- """Mock filepath extraction requests."""
123
- if not settings.enable_filepath_extraction_mock:
124
- return None
125
-
126
- is_fp, cmd, output = is_filepath_extraction_request(request_data)
127
- if not is_fp:
128
- return None
129
-
130
- filepaths = extract_filepaths_from_command(cmd, output)
131
- logger.info("Optimization: Mocked filepath extraction")
132
- return _text_response(
133
- request_data,
134
- filepaths,
135
- input_tokens=100,
136
- output_tokens=10,
137
- )
138
-
139
-
140
- # Cheapest/most common optimizations first for faster short-circuit.
141
- OPTIMIZATION_HANDLERS = [
142
- try_quota_mock,
143
- try_prefix_detection,
144
- try_title_skip,
145
- try_suggestion_skip,
146
- try_filepath_mock,
147
- ]
148
-
149
-
150
- def try_optimizations(
151
- request_data: MessagesRequest,
152
- settings: Settings,
153
- *,
154
- response_model: str | None = None,
155
- ) -> MessagesResponse | None:
156
- """Run optimization handlers in order. Returns first match or None."""
157
- for handler in OPTIMIZATION_HANDLERS:
158
- result = handler(request_data, settings)
159
- if result is not None:
160
- if response_model is None:
161
- return result
162
- return result.model_copy(update={"model": response_model})
163
- return None
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
src/free_claude_code/api/ports.py CHANGED
@@ -53,6 +53,12 @@ class AdminRuntimePort(Protocol):
53
 
54
  async def refresh_claude_vscode(self) -> JsonObject: ...
55
 
 
 
 
 
 
 
56
  async def dsh_desktop_status(self) -> JsonObject: ...
57
 
58
  async def connect_dsh_desktop(self) -> JsonObject: ...
 
53
 
54
  async def refresh_claude_vscode(self) -> JsonObject: ...
55
 
56
+ async def antigravity_status(self) -> JsonObject: ...
57
+
58
+ async def connect_antigravity(self) -> JsonObject: ...
59
+
60
+ async def disconnect_antigravity(self) -> JsonObject: ...
61
+
62
  async def dsh_desktop_status(self) -> JsonObject: ...
63
 
64
  async def connect_dsh_desktop(self) -> JsonObject: ...
src/free_claude_code/api/request_errors.py CHANGED
@@ -18,6 +18,7 @@ from free_claude_code.core.diagnostics import (
18
  redacted_exception_traceback,
19
  safe_exception_message,
20
  )
 
21
  from free_claude_code.core.openai_responses import (
22
  openai_error_payload,
23
  openai_error_type_for_failure,
@@ -27,7 +28,7 @@ from free_claude_code.core.request_outcomes import (
27
  record_request_failure,
28
  )
29
 
30
- WireApi = Literal["messages", "responses"]
31
 
32
 
33
  def require_non_empty_messages(messages: Sequence[object]) -> None:
@@ -43,6 +44,11 @@ def ordinary_application_error_response(
43
  ) -> JSONResponse:
44
  """Serialize a deterministic application error without terminal headers."""
45
  record_request_failure(error.kind.value)
 
 
 
 
 
46
  if wire_api == "responses":
47
  return JSONResponse(
48
  status_code=error.status_code,
 
18
  redacted_exception_traceback,
19
  safe_exception_message,
20
  )
21
+ from free_claude_code.core.gemini import gemini_error_payload
22
  from free_claude_code.core.openai_responses import (
23
  openai_error_payload,
24
  openai_error_type_for_failure,
 
28
  record_request_failure,
29
  )
30
 
31
+ WireApi = Literal["messages", "responses", "gemini"]
32
 
33
 
34
  def require_non_empty_messages(messages: Sequence[object]) -> None:
 
44
  ) -> JSONResponse:
45
  """Serialize a deterministic application error without terminal headers."""
46
  record_request_failure(error.kind.value)
47
+ if wire_api == "gemini":
48
+ return JSONResponse(
49
+ status_code=error.status_code,
50
+ content=gemini_error_payload(error.status_code, error.message),
51
+ )
52
  if wire_api == "responses":
53
  return JSONResponse(
54
  status_code=error.status_code,
src/free_claude_code/api/request_ids.py CHANGED
@@ -10,6 +10,8 @@ from starlette.types import ASGIApp, Message, Receive, Scope, Send
10
 
11
  from free_claude_code.core.trace import extract_claude_session_id_from_headers
12
 
 
 
13
  REQUEST_ID_HEADER = "request-id"
14
  OPENAI_REQUEST_ID_HEADER = "x-request-id"
15
  _REQUEST_ID_STATE_ATTRIBUTE = "fcc_request_id"
@@ -97,5 +99,5 @@ def _set_request_id_headers(
97
  path: str,
98
  ) -> None:
99
  headers[REQUEST_ID_HEADER] = request_id
100
- if path in _OPENAI_REQUEST_ID_PATHS:
101
  headers[OPENAI_REQUEST_ID_HEADER] = request_id
 
10
 
11
  from free_claude_code.core.trace import extract_claude_session_id_from_headers
12
 
13
+ from .generation_paths import generation_api
14
+
15
  REQUEST_ID_HEADER = "request-id"
16
  OPENAI_REQUEST_ID_HEADER = "x-request-id"
17
  _REQUEST_ID_STATE_ATTRIBUTE = "fcc_request_id"
 
99
  path: str,
100
  ) -> None:
101
  headers[REQUEST_ID_HEADER] = request_id
102
+ if path in _OPENAI_REQUEST_ID_PATHS or generation_api(path) == "gemini":
103
  headers[OPENAI_REQUEST_ID_HEADER] = request_id
src/free_claude_code/api/request_lifetime.py CHANGED
@@ -5,6 +5,8 @@ import asyncio
5
 
6
  from starlette.types import ASGIApp, Message, Receive, Scope, Send
7
 
 
 
8
  _CLIENT_OWNED_PATHS = frozenset(
9
  {
10
  "/v1/messages",
@@ -85,9 +87,12 @@ class ClientRequestLifetimeMiddleware:
85
 
86
 
87
  def _owns_client_lifetime(scope: Scope) -> bool:
88
- path = scope.get("path")
89
  return scope["type"] == "http" and (
90
- (scope.get("method") == "POST" and path in _CLIENT_OWNED_PATHS)
 
 
 
91
  or (
92
  scope.get("method") == "GET" and path in {"/v1/models", "/muse-code/models"}
93
  )
 
5
 
6
  from starlette.types import ASGIApp, Message, Receive, Scope, Send
7
 
8
+ from .generation_paths import generation_api
9
+
10
  _CLIENT_OWNED_PATHS = frozenset(
11
  {
12
  "/v1/messages",
 
87
 
88
 
89
  def _owns_client_lifetime(scope: Scope) -> bool:
90
+ path = scope.get("path", "")
91
  return scope["type"] == "http" and (
92
+ (
93
+ scope.get("method") == "POST"
94
+ and (path in _CLIENT_OWNED_PATHS or generation_api(path) is not None)
95
+ )
96
  or (
97
  scope.get("method") == "GET" and path in {"/v1/models", "/muse-code/models"}
98
  )
src/free_claude_code/api/request_outcomes.py CHANGED
@@ -18,11 +18,17 @@ from free_claude_code.core.request_outcomes import (
18
  record_request_failure,
19
  )
20
 
 
 
21
  _FAILURE_EVENTS = frozenset({"error", "response.error", "response.failed"})
22
 
23
 
24
  def _observe_event(event: SSEEvent) -> None:
25
- kind = event.event or event.data.get("type")
 
 
 
 
26
  if not isinstance(kind, str) or kind not in _FAILURE_EVENTS:
27
  return
28
  response = event.data.get("response")
@@ -43,7 +49,7 @@ class RequestOutcomeMiddleware:
43
  if (
44
  scope["type"] != "http"
45
  or scope.get("method") != "POST"
46
- or scope.get("path") not in {"/v1/messages", "/v1/responses"}
47
  ):
48
  await self._app(scope, receive, send)
49
  return
@@ -76,7 +82,11 @@ class RequestOutcomeMiddleware:
76
  .get("content-type", "")
77
  .startswith("text/event-stream")
78
  ):
79
- decoder = AnthropicSSEDecoder(event_names=_FAILURE_EVENTS)
 
 
 
 
80
  elif message["type"] == "http.response.body":
81
  completed = not message.get("more_body", False)
82
  if decoder is not None:
@@ -115,9 +125,7 @@ class RequestOutcomeMiddleware:
115
  result = "success"
116
  logger.bind(
117
  event="request.completed",
118
- wire_api="responses"
119
- if scope["path"] == "/v1/responses"
120
- else "messages",
121
  provider_id=outcome.provider_id,
122
  model=outcome.model,
123
  status_code=status_code,
 
18
  record_request_failure,
19
  )
20
 
21
+ from .generation_paths import generation_api
22
+
23
  _FAILURE_EVENTS = frozenset({"error", "response.error", "response.failed"})
24
 
25
 
26
  def _observe_event(event: SSEEvent) -> None:
27
+ kind = (
28
+ event.event
29
+ or event.data.get("type")
30
+ or ("error" if "error" in event.data else None)
31
+ )
32
  if not isinstance(kind, str) or kind not in _FAILURE_EVENTS:
33
  return
34
  response = event.data.get("response")
 
49
  if (
50
  scope["type"] != "http"
51
  or scope.get("method") != "POST"
52
+ or generation_api(scope.get("path", "")) is None
53
  ):
54
  await self._app(scope, receive, send)
55
  return
 
82
  .get("content-type", "")
83
  .startswith("text/event-stream")
84
  ):
85
+ decoder = AnthropicSSEDecoder(
86
+ event_names=None
87
+ if generation_api(scope["path"]) == "gemini"
88
+ else _FAILURE_EVENTS
89
+ )
90
  elif message["type"] == "http.response.body":
91
  completed = not message.get("more_body", False)
92
  if decoder is not None:
 
125
  result = "success"
126
  logger.bind(
127
  event="request.completed",
128
+ wire_api=generation_api(scope["path"]),
 
 
129
  provider_id=outcome.provider_id,
130
  model=outcome.model,
131
  status_code=status_code,
src/free_claude_code/api/response_streams.py CHANGED
@@ -25,6 +25,8 @@ from free_claude_code.core.anthropic.streaming import (
25
  from free_claude_code.core.async_iterators import try_close_async_iterator
26
  from free_claude_code.core.diagnostics import safe_exception_message
27
  from free_claude_code.core.failures import find_execution_failure
 
 
28
  from free_claude_code.core.json_types import JsonObject
29
  from free_claude_code.core.openai_responses import (
30
  committed_response_failure_frame,
@@ -45,7 +47,7 @@ PreStartErrorResponse = Callable[[BaseException], Response]
45
  TerminalFrameEmitter = Callable[[str, str, BaseException], str]
46
  TerminalFailureObserver = Callable[[BaseException], None]
47
  ReleaseResponseResource = Callable[[], Awaitable[None]]
48
- WireApi = Literal["messages", "responses"]
49
 
50
 
51
  class EmptyStreamError(RuntimeError):
@@ -394,6 +396,41 @@ async def openai_responses_sse_streaming_response(
394
  )
395
 
396
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
397
  def _trace_responses_terminal_failure(
398
  exc: BaseException,
399
  *,
 
25
  from free_claude_code.core.async_iterators import try_close_async_iterator
26
  from free_claude_code.core.diagnostics import safe_exception_message
27
  from free_claude_code.core.failures import find_execution_failure
28
+ from free_claude_code.core.gemini import gemini_error_payload
29
+ from free_claude_code.core.gemini.response import gemini_frame
30
  from free_claude_code.core.json_types import JsonObject
31
  from free_claude_code.core.openai_responses import (
32
  committed_response_failure_frame,
 
47
  TerminalFrameEmitter = Callable[[str, str, BaseException], str]
48
  TerminalFailureObserver = Callable[[BaseException], None]
49
  ReleaseResponseResource = Callable[[], Awaitable[None]]
50
+ WireApi = Literal["messages", "responses", "gemini"]
51
 
52
 
53
  class EmptyStreamError(RuntimeError):
 
396
  )
397
 
398
 
399
+ async def gemini_sse_streaming_response(
400
+ body: AsyncIterator[str],
401
+ *,
402
+ pre_start_error_response: PreStartErrorResponse,
403
+ request_id: str,
404
+ ) -> Response:
405
+ def terminal_frame(_first: str, _latest: str, error: BaseException) -> str:
406
+ failure = find_execution_failure(error)
407
+ return gemini_frame(
408
+ gemini_error_payload(
409
+ failure.status_code if failure else 500,
410
+ safe_exception_message(failure or error),
411
+ )
412
+ )
413
+
414
+ def terminal_observer(error: BaseException) -> None:
415
+ failure = find_execution_failure(error)
416
+ trace_terminal_execution_error(
417
+ wire_api="gemini",
418
+ request_id=request_id,
419
+ status_code=failure.status_code if failure else 500,
420
+ error_type=failure.kind.value if failure else "api_error",
421
+ error=error,
422
+ )
423
+
424
+ return await _first_chunk_streaming_response(
425
+ body,
426
+ wire_api="gemini",
427
+ headers={"Cache-Control": "no-cache", "X-Accel-Buffering": "no"},
428
+ pre_start_error_response=pre_start_error_response,
429
+ terminal_frame=terminal_frame,
430
+ terminal_failure_observer=terminal_observer,
431
+ )
432
+
433
+
434
  def _trace_responses_terminal_failure(
435
  exc: BaseException,
436
  *,
src/free_claude_code/api/routes.py CHANGED
@@ -1,6 +1,7 @@
1
  from __future__ import annotations
2
  """FastAPI route handlers."""
3
 
 
4
  from collections.abc import Mapping
5
  from typing import Annotated, Any, cast
6
 
@@ -25,6 +26,7 @@ from free_claude_code.core.anthropic.native import (
25
  validate_messages_json,
26
  )
27
  from free_claude_code.core.anthropic.passthrough import NativeMessagesRequest
 
28
  from free_claude_code.core.json_types import JsonObject
29
  from free_claude_code.core.openai_responses import OpenAIResponsesRequest
30
  from free_claude_code.core.trace import trace_event
@@ -33,10 +35,12 @@ from .dependencies import (
33
  get_services,
34
  get_settings,
35
  require_anthropic_proxy_auth,
 
36
  require_proxy_auth,
37
  resolve_provider,
38
  )
39
  from .handlers import MessagesHandler, ResponsesHandler, TokenCountHandler
 
40
  from .model_catalog import (
41
  ModelCatalogView,
42
  ModelsListResponse,
@@ -51,6 +55,78 @@ from .response_streams import bind_response_lifetime
51
  router = APIRouter()
52
 
53
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
54
  def _provider_resolver(lease: RequestRuntimeLease) -> ProviderResolver:
55
  return lambda provider_type, api_key_number: resolve_provider(
56
  provider_type, api_key_number, lease=lease
@@ -320,7 +396,7 @@ async def list_models(
320
  ):
321
  """List the model ids this proxy advertises to compatible clients."""
322
  trace_event(stage="ingress", event="free_claude_code.api.models.list", source="api")
323
- snapshot = await services.requests.wait_for_catalog()
324
  return build_models_list_response(
325
  snapshot.settings,
326
  snapshot,
@@ -339,7 +415,7 @@ async def list_muse_models(
339
  ):
340
  """List the direct Responses models expected by Muse Code."""
341
  trace_event(stage="ingress", event="free_claude_code.api.models.list", source="api")
342
- snapshot = await services.requests.wait_for_catalog()
343
  return build_muse_models_list_response(snapshot.settings, snapshot)
344
 
345
 
 
1
  from __future__ import annotations
2
  """FastAPI route handlers."""
3
 
4
+ import json
5
  from collections.abc import Mapping
6
  from typing import Annotated, Any, cast
7
 
 
26
  validate_messages_json,
27
  )
28
  from free_claude_code.core.anthropic.passthrough import NativeMessagesRequest
29
+ from free_claude_code.core.gemini import GeminiRequest
30
  from free_claude_code.core.json_types import JsonObject
31
  from free_claude_code.core.openai_responses import OpenAIResponsesRequest
32
  from free_claude_code.core.trace import trace_event
 
35
  get_services,
36
  get_settings,
37
  require_anthropic_proxy_auth,
38
+ require_gemini_proxy_auth,
39
  require_proxy_auth,
40
  resolve_provider,
41
  )
42
  from .handlers import MessagesHandler, ResponsesHandler, TokenCountHandler
43
+ from .handlers.gemini import GeminiHandler
44
  from .model_catalog import (
45
  ModelCatalogView,
46
  ModelsListResponse,
 
55
  router = APIRouter()
56
 
57
 
58
+ def _unique_json_fields(pairs: list[tuple[str, Any]]) -> dict[str, Any]:
59
+ result: dict[str, Any] = {}
60
+ for key, value in pairs:
61
+ if key in result:
62
+ raise ValueError(f"Duplicate JSON field: {key}")
63
+ result[key] = value
64
+ return result
65
+
66
+
67
+ @router.post("/v1beta/models/{resource:path}")
68
+ async def create_gemini_response(
69
+ resource: str,
70
+ request: Request,
71
+ request_data: Annotated[dict[str, Any], Body()],
72
+ services: ApiServices = Depends(get_services),
73
+ _auth=Depends(require_gemini_proxy_auth),
74
+ ):
75
+ try:
76
+ json.loads(await request.body(), object_pairs_hook=_unique_json_fields)
77
+ except ValueError as error:
78
+ raise HTTPException(status_code=400, detail=str(error)) from error
79
+ model, separator, action = resource.rpartition(":")
80
+ if not separator or action not in {"generateContent", "streamGenerateContent"}:
81
+ raise HTTPException(status_code=404, detail="Unknown Gemini generation method")
82
+ if not model.strip() or any(
83
+ not segment or segment in {".", ".."} or any(char.isspace() for char in segment)
84
+ for segment in model.split("/")
85
+ ):
86
+ raise HTTPException(status_code=400, detail="Invalid model resource")
87
+ streaming = action == "streamGenerateContent"
88
+ alt = request.query_params.getlist("alt")
89
+ if (
90
+ len(alt) > 1
91
+ or (streaming and alt != ["sse"])
92
+ or (not streaming and alt not in ([], ["json"]))
93
+ ):
94
+ raise HTTPException(
95
+ status_code=400,
96
+ detail="Streaming requires alt=sse; JSON generation accepts alt=json",
97
+ )
98
+ if "model" in request_data:
99
+ raise HTTPException(status_code=400, detail="Specify the model in the URL only")
100
+ try:
101
+ data = GeminiRequest.model_validate({**request_data, "model": model})
102
+ except ValidationError as error:
103
+ raise RequestValidationError(
104
+ [{**item, "loc": ("body", *item["loc"])} for item in error.errors()],
105
+ body=request_data,
106
+ ) from error
107
+ lease = await services.requests.acquire()
108
+ try:
109
+ await lease.wait_for_token_estimation()
110
+ handler = GeminiHandler(
111
+ lease.settings,
112
+ _provider_resolver(lease),
113
+ generation_id=lease.generation_id,
114
+ request_headers={
115
+ key: value
116
+ for key, value in request.headers.items()
117
+ if key.lower() not in {"authorization", "x-api-key", "x-goog-api-key"}
118
+ },
119
+ model_info_lookup=lease.model_info,
120
+ )
121
+ response = await handler.create(
122
+ data, streaming=streaming, request_id=get_request_id(request)
123
+ )
124
+ except BaseException:
125
+ await lease.release()
126
+ raise
127
+ return await bind_response_lifetime(response, lease.release)
128
+
129
+
130
  def _provider_resolver(lease: RequestRuntimeLease) -> ProviderResolver:
131
  return lambda provider_type, api_key_number: resolve_provider(
132
  provider_type, api_key_number, lease=lease
 
396
  ):
397
  """List the model ids this proxy advertises to compatible clients."""
398
  trace_event(stage="ingress", event="free_claude_code.api.models.list", source="api")
399
+ snapshot = services.requests.catalog_snapshot()
400
  return build_models_list_response(
401
  snapshot.settings,
402
  snapshot,
 
415
  ):
416
  """List the direct Responses models expected by Muse Code."""
417
  trace_event(stage="ingress", event="free_claude_code.api.models.list", source="api")
418
+ snapshot = services.requests.catalog_snapshot()
419
  return build_muse_models_list_response(snapshot.settings, snapshot)
420
 
421
 
src/free_claude_code/api/stream_delivery.py CHANGED
@@ -9,23 +9,14 @@ from typing import Any, Literal
9
  try:
10
  import simplejson
11
  except ImportError:
12
- import json
13
- class _SimpleJsonFallback:
14
- @staticmethod
15
- def dumps(obj, *args, **kwargs):
16
- kwargs.pop('use_decimal', None)
17
- return json.dumps(obj, *args, **kwargs)
18
- @staticmethod
19
- def loads(s, *args, **kwargs):
20
- kwargs.pop('use_decimal', None)
21
- return json.loads(s, *args, **kwargs)
22
- simplejson = _SimpleJsonFallback()
23
 
24
  from free_claude_code.core.anthropic.streaming.decoder import AnthropicSSEDecoder
25
  from free_claude_code.core.async_iterators import try_close_async_iterator
26
  from free_claude_code.core.continuation_stream import ContinuationStream
27
  from free_claude_code.core.delivered_response import DeliveredResponse
28
  from free_claude_code.core.failures import find_execution_failure
 
29
  from free_claude_code.core.stream_delivery import (
30
  StreamDeliveryState,
31
  bind_stream_delivery,
@@ -65,7 +56,7 @@ class PublicResponseStream(AsyncIterator[str]):
65
  body: AsyncIterator[str],
66
  state: StreamDeliveryState,
67
  *,
68
- wire_api: Literal["messages", "responses"],
69
  first_chunk: str,
70
  terminal_frame: Callable[[str, str, BaseException], str] | None,
71
  terminal_failure_observer: Callable[[BaseException], None] | None,
@@ -239,6 +230,8 @@ class PublicResponseStream(AsyncIterator[str]):
239
 
240
  @staticmethod
241
  def _format(payload: dict[str, Any]) -> str:
 
 
242
  return f"event: {payload['type']}\ndata: {simplejson.dumps(payload, use_decimal=True)}\n\n"
243
 
244
  async def aclose(self) -> None:
@@ -289,6 +282,13 @@ class PublicResponseStream(AsyncIterator[str]):
289
  return frame
290
 
291
  raw_kind = payload.get("type")
 
 
 
 
 
 
 
292
  kind = raw_kind if isinstance(raw_kind, str) else ""
293
  metadata = self._metadata(kind, payload)
294
  if self._wire_api == "messages":
@@ -358,6 +358,12 @@ class PublicResponseStream(AsyncIterator[str]):
358
  return self._release_content(frame, synthetic=synthetic)
359
 
360
  def _metadata(self, kind: str, payload: dict[str, Any]) -> bool:
 
 
 
 
 
 
361
  if kind == "ping":
362
  return True
363
  if self._wire_api == "messages" and kind == "message_start":
 
9
  try:
10
  import simplejson
11
  except ImportError:
12
+ import json as simplejson
 
 
 
 
 
 
 
 
 
 
13
 
14
  from free_claude_code.core.anthropic.streaming.decoder import AnthropicSSEDecoder
15
  from free_claude_code.core.async_iterators import try_close_async_iterator
16
  from free_claude_code.core.continuation_stream import ContinuationStream
17
  from free_claude_code.core.delivered_response import DeliveredResponse
18
  from free_claude_code.core.failures import find_execution_failure
19
+ from free_claude_code.core.gemini.response import gemini_finished, gemini_parts
20
  from free_claude_code.core.stream_delivery import (
21
  StreamDeliveryState,
22
  bind_stream_delivery,
 
56
  body: AsyncIterator[str],
57
  state: StreamDeliveryState,
58
  *,
59
+ wire_api: Literal["messages", "responses", "gemini"],
60
  first_chunk: str,
61
  terminal_frame: Callable[[str, str, BaseException], str] | None,
62
  terminal_failure_observer: Callable[[BaseException], None] | None,
 
230
 
231
  @staticmethod
232
  def _format(payload: dict[str, Any]) -> str:
233
+ if "type" not in payload:
234
+ return f"data: {simplejson.dumps(payload, use_decimal=True)}\n\n"
235
  return f"event: {payload['type']}\ndata: {simplejson.dumps(payload, use_decimal=True)}\n\n"
236
 
237
  async def aclose(self) -> None:
 
282
  return frame
283
 
284
  raw_kind = payload.get("type")
285
+ if self._wire_api == "gemini":
286
+ if self._public_id is None and isinstance(payload.get("responseId"), str):
287
+ self._public_id = payload["responseId"]
288
+ self.start_frame = frame
289
+ elif self._public_id is not None and "responseId" in payload:
290
+ payload["responseId"] = self._public_id
291
+ frame = replace_frame_data(frame, payload)
292
  kind = raw_kind if isinstance(raw_kind, str) else ""
293
  metadata = self._metadata(kind, payload)
294
  if self._wire_api == "messages":
 
358
  return self._release_content(frame, synthetic=synthetic)
359
 
360
  def _metadata(self, kind: str, payload: dict[str, Any]) -> bool:
361
+ if self._wire_api == "gemini":
362
+ return (
363
+ not gemini_parts(payload)
364
+ and not gemini_finished(payload)
365
+ and "error" not in payload
366
+ )
367
  if kind == "ping":
368
  return True
369
  if self._wire_api == "messages" and kind == "message_start":
src/free_claude_code/api/stream_frames.py CHANGED
@@ -8,17 +8,7 @@ from typing import Any
8
  try:
9
  import simplejson
10
  except ImportError:
11
- import json
12
- class _SimpleJsonFallback:
13
- @staticmethod
14
- def dumps(obj, *args, **kwargs):
15
- kwargs.pop('use_decimal', None)
16
- return json.dumps(obj, *args, **kwargs)
17
- @staticmethod
18
- def loads(s, *args, **kwargs):
19
- kwargs.pop('use_decimal', None)
20
- return json.loads(s, *args, **kwargs)
21
- simplejson = _SimpleJsonFallback()
22
 
23
 
24
  def frame_data(frame: str) -> dict[str, Any] | None:
 
8
  try:
9
  import simplejson
10
  except ImportError:
11
+ import json as simplejson
 
 
 
 
 
 
 
 
 
 
12
 
13
 
14
  def frame_data(frame: str) -> dict[str, Any] | None:
src/free_claude_code/api/tool_call_buffer.py CHANGED
@@ -48,7 +48,7 @@ def _client_call(item: Mapping[str, object]) -> bool:
48
  class ToolCallBuffer:
49
  """Select complete groups in source order, without owning public delivery."""
50
 
51
- def __init__(self, wire_api: Literal["messages", "responses"]) -> None:
52
  self._wire_api = wire_api
53
  self._frames: list[str] = []
54
  self._calls: set[CallKey] = set()
@@ -59,6 +59,9 @@ class ToolCallBuffer:
59
  self._indices: dict[int, CallKey] = {}
60
 
61
  def feed(self, frame: str) -> list[str]:
 
 
 
62
  events = parse_sse_text(frame)
63
  if not events:
64
  return self._publish(frame)
 
48
  class ToolCallBuffer:
49
  """Select complete groups in source order, without owning public delivery."""
50
 
51
+ def __init__(self, wire_api: Literal["messages", "responses", "gemini"]) -> None:
52
  self._wire_api = wire_api
53
  self._frames: list[str] = []
54
  self._calls: set[CallKey] = set()
 
59
  self._indices: dict[int, CallKey] = {}
60
 
61
  def feed(self, frame: str) -> list[str]:
62
+ if self._wire_api == "gemini":
63
+ # Gemini presenters release only complete, object-valued calls.
64
+ return [frame]
65
  events = parse_sse_text(frame)
66
  if not events:
67
  return self._publish(frame)
src/free_claude_code/application/code_sessions/service.py CHANGED
@@ -3,7 +3,7 @@ from __future__ import annotations
3
 
4
  import asyncio
5
  import uuid
6
- from collections.abc import AsyncIterator, Callable, Coroutine, Sequence
7
  from contextlib import asynccontextmanager
8
  from dataclasses import dataclass, field
9
  from pathlib import Path
@@ -252,7 +252,7 @@ class CodeService:
252
  del self._gates[session_id]
253
 
254
  @asynccontextmanager
255
- async def _locked(self, owner: _SessionRuntime) -> AsyncIterator[None]:
256
  async with owner.lock:
257
  try:
258
  yield
 
3
 
4
  import asyncio
5
  import uuid
6
+ from collections.abc import AsyncGenerator, Callable, Coroutine, Sequence
7
  from contextlib import asynccontextmanager
8
  from dataclasses import dataclass, field
9
  from pathlib import Path
 
252
  del self._gates[session_id]
253
 
254
  @asynccontextmanager
255
+ async def _locked(self, owner: _SessionRuntime) -> AsyncGenerator[None]:
256
  async with owner.lock:
257
  try:
258
  yield
src/free_claude_code/application/execution.py CHANGED
@@ -18,6 +18,7 @@ from free_claude_code.core.anthropic import (
18
  )
19
  from free_claude_code.core.anthropic.tokens import TokenCounter
20
  from free_claude_code.core.failures import ExecutionFailure, FailureKind
 
21
  from free_claude_code.core.openai_responses import (
22
  OpenAIResponsesRequest,
23
  estimate_responses_input_tokens,
@@ -47,13 +48,14 @@ from .routing import (
47
  ProviderModelCandidate,
48
  ProviderModelTarget,
49
  ResolvedModelRoute,
 
50
  RoutedMessagesRequest,
51
  RoutedNativeMessagesRequest,
52
  RoutedResponsesRequest,
53
  )
54
 
55
  ResponsesTokenCounter = Callable[[OpenAIResponsesRequest], int]
56
- WireApi = Literal["messages", "responses"]
57
  CandidateStreamOpener = Callable[
58
  [int, ProviderModelCandidate, ContinuationSeed | None],
59
  Awaitable[AsyncIterator[str]],
@@ -371,6 +373,63 @@ class ProviderExecutor:
371
  open_candidate=open_candidate,
372
  )
373
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
374
  def _stream_candidates(
375
  self,
376
  *,
@@ -406,8 +465,8 @@ class ProviderExecutor:
406
  ),
407
  "reasoning_budget_tokens": reasoning.budget_tokens,
408
  }
409
- if wire_api == "responses":
410
- route_trace["wire_api"] = "responses"
411
  if self._generation_id is not None:
412
  route_trace["generation_id"] = self._generation_id
413
  trace_event(**route_trace)
 
18
  )
19
  from free_claude_code.core.anthropic.tokens import TokenCounter
20
  from free_claude_code.core.failures import ExecutionFailure, FailureKind
21
+ from free_claude_code.core.gemini.tokens import estimate_gemini_input_tokens
22
  from free_claude_code.core.openai_responses import (
23
  OpenAIResponsesRequest,
24
  estimate_responses_input_tokens,
 
48
  ProviderModelCandidate,
49
  ProviderModelTarget,
50
  ResolvedModelRoute,
51
+ RoutedGeminiRequest,
52
  RoutedMessagesRequest,
53
  RoutedNativeMessagesRequest,
54
  RoutedResponsesRequest,
55
  )
56
 
57
  ResponsesTokenCounter = Callable[[OpenAIResponsesRequest], int]
58
+ WireApi = Literal["messages", "responses", "gemini"]
59
  CandidateStreamOpener = Callable[
60
  [int, ProviderModelCandidate, ContinuationSeed | None],
61
  Awaitable[AsyncIterator[str]],
 
373
  open_candidate=open_candidate,
374
  )
375
 
376
+ def stream_gemini(
377
+ self,
378
+ routed: RoutedGeminiRequest,
379
+ *,
380
+ raw_log_payload: Callable[[], object],
381
+ request_id: str,
382
+ ) -> AsyncIterator[str]:
383
+ client_reasoning = routed.request.reasoning_policy()
384
+ input_tokens = estimate_gemini_input_tokens(routed.request)
385
+
386
+ async def open_candidate(
387
+ index: int,
388
+ candidate: ProviderModelCandidate,
389
+ continuation: ContinuationSeed | None,
390
+ ) -> AsyncIterator[str]:
391
+ target = candidate.target
392
+ provider = await self._provider_resolver(
393
+ target.provider_id, target.api_key_number
394
+ )
395
+ request = prepare_fallback_request(
396
+ routed.request,
397
+ model=target.provider_model,
398
+ preference=candidate.reasoning_preference,
399
+ )
400
+ return provider.stream_gemini(
401
+ request,
402
+ input_tokens=input_tokens,
403
+ request_id=request_id,
404
+ response_model=routed.resolved.original_model,
405
+ reasoning=resolve_fallback_reasoning(
406
+ routed.reasoning,
407
+ candidate.reasoning_preference,
408
+ client=client_reasoning,
409
+ ),
410
+ request_headers=self._request_headers,
411
+ continuation=continuation,
412
+ model_info=self._model_info_lookup(
413
+ target.provider_id, target.provider_model
414
+ ),
415
+ )
416
+
417
+ return self._stream_candidates(
418
+ resolved=routed.resolved,
419
+ reasoning=routed.reasoning,
420
+ wire_api="gemini",
421
+ raw_log_label="FULL_GEMINI_PAYLOAD",
422
+ raw_log_payload=raw_log_payload,
423
+ request_snapshot=lambda: {
424
+ "model": routed.request.model,
425
+ "content_count": len(routed.request.contents),
426
+ },
427
+ ingress_count_name="content_count",
428
+ ingress_count=len(routed.request.contents),
429
+ request_id=request_id,
430
+ open_candidate=open_candidate,
431
+ )
432
+
433
  def _stream_candidates(
434
  self,
435
  *,
 
465
  ),
466
  "reasoning_budget_tokens": reasoning.budget_tokens,
467
  }
468
+ if wire_api != "messages":
469
+ route_trace["wire_api"] = wire_api
470
  if self._generation_id is not None:
471
  route_trace["generation_id"] = self._generation_id
472
  trace_event(**route_trace)
src/free_claude_code/application/model_metadata.py CHANGED
@@ -1,7 +1,7 @@
1
  from __future__ import annotations
2
  """Application-owned model metadata."""
3
 
4
- from dataclasses import dataclass
5
 
6
  from free_claude_code.core.model_capabilities import ModelInputModality
7
  from free_claude_code.core.reasoning import ReasoningCapability
@@ -25,3 +25,4 @@ class ProviderModelRefreshResult:
25
 
26
  refreshed_provider_ids: tuple[str, ...] = ()
27
  failed_provider_ids: tuple[str, ...] = ()
 
 
1
  from __future__ import annotations
2
  """Application-owned model metadata."""
3
 
4
+ from dataclasses import dataclass, field
5
 
6
  from free_claude_code.core.model_capabilities import ModelInputModality
7
  from free_claude_code.core.reasoning import ReasoningCapability
 
25
 
26
  refreshed_provider_ids: tuple[str, ...] = ()
27
  failed_provider_ids: tuple[str, ...] = ()
28
+ failure_messages: dict[str, str] = field(default_factory=dict)
src/free_claude_code/application/ports.py CHANGED
@@ -8,6 +8,7 @@ from typing import Protocol
8
  from free_claude_code.config.settings import Settings
9
  from free_claude_code.core.anthropic import MessagesRequest
10
  from free_claude_code.core.anthropic.passthrough import NativeMessagesRequest
 
11
  from free_claude_code.core.json_types import JsonObject
12
  from free_claude_code.core.openai_responses import OpenAIResponsesRequest
13
  from free_claude_code.core.reasoning import ReasoningPolicy
@@ -19,6 +20,19 @@ from .model_metadata import ProviderModelInfo
19
  class ProviderPort(Protocol):
20
  """Minimal provider capability required to execute one request."""
21
 
 
 
 
 
 
 
 
 
 
 
 
 
 
22
  def stream_native_messages(
23
  self,
24
  request: NativeMessagesRequest,
@@ -127,7 +141,9 @@ class RequestRuntimePort(ModelCatalogPort, Protocol):
127
 
128
  async def acquire(self) -> RequestRuntimeLease: ...
129
 
130
- async def wait_for_catalog(self) -> ModelCatalogSnapshot: ...
 
 
131
 
132
  def catalog_status(self) -> JsonObject: ...
133
 
 
8
  from free_claude_code.config.settings import Settings
9
  from free_claude_code.core.anthropic import MessagesRequest
10
  from free_claude_code.core.anthropic.passthrough import NativeMessagesRequest
11
+ from free_claude_code.core.gemini import GeminiRequest
12
  from free_claude_code.core.json_types import JsonObject
13
  from free_claude_code.core.openai_responses import OpenAIResponsesRequest
14
  from free_claude_code.core.reasoning import ReasoningPolicy
 
20
  class ProviderPort(Protocol):
21
  """Minimal provider capability required to execute one request."""
22
 
23
+ def stream_gemini(
24
+ self,
25
+ request: GeminiRequest,
26
+ *,
27
+ input_tokens: int,
28
+ request_id: str,
29
+ response_model: str,
30
+ reasoning: ReasoningPolicy,
31
+ request_headers: Mapping[str, str] | None = None,
32
+ model_info: ProviderModelInfo | None = None,
33
+ continuation: ContinuationSeed | None = None,
34
+ ) -> AsyncIterator[str]: ...
35
+
36
  def stream_native_messages(
37
  self,
38
  request: NativeMessagesRequest,
 
141
 
142
  async def acquire(self) -> RequestRuntimeLease: ...
143
 
144
+ def catalog_snapshot(self) -> ModelCatalogSnapshot: ...
145
+
146
+ async def wait_for_model_catalog(self, model: str) -> ModelCatalogSnapshot: ...
147
 
148
  def catalog_status(self) -> JsonObject: ...
149
 
src/free_claude_code/application/reasoning.py CHANGED
@@ -5,6 +5,7 @@ from collections.abc import Mapping
5
 
6
  from free_claude_code.config.reasoning import ReasoningPreference
7
  from free_claude_code.core.anthropic.models import MessagesRequest, ThinkingConfig
 
8
  from free_claude_code.core.openai_responses import (
9
  OpenAIResponsesRequest,
10
  responses_reasoning_policy,
@@ -58,9 +59,16 @@ def resolve_fallback_reasoning(
58
  return _apply_reasoning_preference(original, preference)
59
 
60
 
61
- def prepare_fallback_request[T: MessagesRequest | OpenAIResponsesRequest](
62
- original: T, *, model: str, preference: ReasoningPreference
63
- ) -> T:
 
 
 
 
 
 
 
64
  """Copy a candidate and remove computation controls replaced by its policy."""
65
 
66
  request = original.model_copy(update={"model": model}, deep=True)
@@ -76,6 +84,10 @@ def prepare_fallback_request[T: MessagesRequest | OpenAIResponsesRequest](
76
  )
77
  if request.output_config is not None:
78
  request.output_config.pop("effort", None)
 
 
 
 
79
  elif request.reasoning is not None:
80
  request.reasoning.pop("effort", None)
81
  return request
 
5
 
6
  from free_claude_code.config.reasoning import ReasoningPreference
7
  from free_claude_code.core.anthropic.models import MessagesRequest, ThinkingConfig
8
+ from free_claude_code.core.gemini import GeminiRequest
9
  from free_claude_code.core.openai_responses import (
10
  OpenAIResponsesRequest,
11
  responses_reasoning_policy,
 
59
  return _apply_reasoning_preference(original, preference)
60
 
61
 
62
+ def resolve_gemini_reasoning_policy(
63
+ request: GeminiRequest,
64
+ preference: ReasoningPreference,
65
+ ) -> ReasoningPolicy:
66
+ return _apply_reasoning_preference(request.reasoning_policy(), preference)
67
+
68
+
69
+ def prepare_fallback_request[
70
+ T: MessagesRequest | OpenAIResponsesRequest | GeminiRequest
71
+ ](original: T, *, model: str, preference: ReasoningPreference) -> T:
72
  """Copy a candidate and remove computation controls replaced by its policy."""
73
 
74
  request = original.model_copy(update={"model": model}, deep=True)
 
84
  )
85
  if request.output_config is not None:
86
  request.output_config.pop("effort", None)
87
+ elif isinstance(request, GeminiRequest):
88
+ thinking = request.generation_config.thinking_config
89
+ thinking.thinking_budget = None
90
+ thinking.thinking_level = None
91
  elif request.reasoning is not None:
92
  request.reasoning.pop("effort", None)
93
  return request
src/free_claude_code/application/routing.py CHANGED
@@ -13,6 +13,7 @@ from free_claude_code.config.model_refs import (
13
  is_retired_model_ref,
14
  parse_model_name,
15
  parse_provider_type,
 
16
  )
17
  from free_claude_code.config.provider_catalog import PROVIDER_CATALOG
18
  from free_claude_code.config.reasoning import ReasoningPreference
@@ -28,11 +29,16 @@ from free_claude_code.core.gateway_model_ids import (
28
  DESKTOP_NO_THINKING_PREFIX,
29
  decode_gateway_model_id,
30
  )
 
31
  from free_claude_code.core.openai_responses import OpenAIResponsesRequest
32
  from free_claude_code.core.reasoning import ReasoningPolicy
33
  from free_claude_code.core.trace import trace_event
34
 
35
- from .reasoning import resolve_reasoning_policy, resolve_responses_reasoning_policy
 
 
 
 
36
 
37
  _ROUTE_SETTINGS = (
38
  ("fable", "model_fable", "reasoning_fable"),
@@ -98,6 +104,13 @@ class RoutedResponsesRequest:
98
  reasoning: ReasoningPolicy
99
 
100
 
 
 
 
 
 
 
 
101
  @dataclass(frozen=True, slots=True)
102
  class RoutedTokenCountRequest:
103
  request: TokenCountRequest | NativeTokenCountRequest
@@ -211,6 +224,13 @@ class ModelRouter:
211
  parse_model_name(self._settings.model),
212
  decoded.force_reasoning_off if decoded is not None else False,
213
  )
 
 
 
 
 
 
 
214
  if decoded is not None:
215
  if decoded.provider_id not in self._settings.provider_ids:
216
  if decoded.provider_id.startswith("custom_") or model_name.partition(
@@ -362,3 +382,15 @@ class ModelRouter:
362
  resolved.reasoning_preference,
363
  ),
364
  )
 
 
 
 
 
 
 
 
 
 
 
 
 
13
  is_retired_model_ref,
14
  parse_model_name,
15
  parse_provider_type,
16
+ replace_retired_model_ref,
17
  )
18
  from free_claude_code.config.provider_catalog import PROVIDER_CATALOG
19
  from free_claude_code.config.reasoning import ReasoningPreference
 
29
  DESKTOP_NO_THINKING_PREFIX,
30
  decode_gateway_model_id,
31
  )
32
+ from free_claude_code.core.gemini import GeminiRequest
33
  from free_claude_code.core.openai_responses import OpenAIResponsesRequest
34
  from free_claude_code.core.reasoning import ReasoningPolicy
35
  from free_claude_code.core.trace import trace_event
36
 
37
+ from .reasoning import (
38
+ resolve_gemini_reasoning_policy,
39
+ resolve_reasoning_policy,
40
+ resolve_responses_reasoning_policy,
41
+ )
42
 
43
  _ROUTE_SETTINGS = (
44
  ("fable", "model_fable", "reasoning_fable"),
 
104
  reasoning: ReasoningPolicy
105
 
106
 
107
+ @dataclass(frozen=True, slots=True)
108
+ class RoutedGeminiRequest:
109
+ request: GeminiRequest
110
+ resolved: ResolvedModelRoute
111
+ reasoning: ReasoningPolicy
112
+
113
+
114
  @dataclass(frozen=True, slots=True)
115
  class RoutedTokenCountRequest:
116
  request: TokenCountRequest | NativeTokenCountRequest
 
224
  parse_model_name(self._settings.model),
225
  decoded.force_reasoning_off if decoded is not None else False,
226
  )
227
+ replacement = replace_retired_model_ref(candidate)
228
+ if replacement != candidate:
229
+ return (
230
+ parse_provider_type(replacement),
231
+ parse_model_name(replacement),
232
+ decoded.force_reasoning_off if decoded is not None else False,
233
+ )
234
  if decoded is not None:
235
  if decoded.provider_id not in self._settings.provider_ids:
236
  if decoded.provider_id.startswith("custom_") or model_name.partition(
 
382
  resolved.reasoning_preference,
383
  ),
384
  )
385
+
386
+ def resolve_gemini_request(self, request: GeminiRequest) -> RoutedGeminiRequest:
387
+ resolved = self.resolve(request.model)
388
+ return RoutedGeminiRequest(
389
+ request=request.model_copy(
390
+ update={"model": resolved.primary.provider_model}, deep=True
391
+ ),
392
+ resolved=resolved,
393
+ reasoning=resolve_gemini_reasoning_policy(
394
+ request, resolved.reasoning_preference
395
+ ),
396
+ )
src/free_claude_code/cli/desktop_entrypoint.py CHANGED
@@ -1,6 +1,7 @@
1
  from __future__ import annotations
2
  """Lightweight entrypoint for the optional FCC desktop shell."""
3
 
 
4
  import sys
5
  from collections.abc import Sequence
6
  from pathlib import Path
@@ -16,6 +17,15 @@ def launch(argv: Sequence[str] | None = None) -> None:
16
  if len(args) == 2 and args[0] == "--export-icon":
17
  export_app_icon(Path(args[1]))
18
  return
 
 
 
 
 
 
 
 
 
19
  auto_started = args == ("--auto-started",)
20
  if args and not auto_started:
21
  print("Usage: fcc-desktop [--export-icon PATH]", file=sys.stderr)
 
1
  from __future__ import annotations
2
  """Lightweight entrypoint for the optional FCC desktop shell."""
3
 
4
+ import subprocess
5
  import sys
6
  from collections.abc import Sequence
7
  from pathlib import Path
 
17
  if len(args) == 2 and args[0] == "--export-icon":
18
  export_app_icon(Path(args[1]))
19
  return
20
+ if args == ("--install-macos-app",) and sys.platform == "darwin":
21
+ from free_claude_code.cli.macos_app import install
22
+
23
+ try:
24
+ install()
25
+ except (OSError, ValueError, subprocess.CalledProcessError) as error:
26
+ print(f"Could not install the macOS app: {error}", file=sys.stderr)
27
+ raise SystemExit(1) from error
28
+ return
29
  auto_started = args == ("--auto-started",)
30
  if args and not auto_started:
31
  print("Usage: fcc-desktop [--export-icon PATH]", file=sys.stderr)
src/free_claude_code/cli/doctor.py CHANGED
@@ -12,10 +12,29 @@ from .clipboard import ClipboardUnavailable, copy_text
12
 
13
 
14
  def main(argv: Sequence[str] | None = None) -> None:
15
- argparse.ArgumentParser(
16
  prog="fcc-doctor",
17
  description="Print and copy FCC diagnostics for a bug report.",
18
- ).parse_args(argv)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
19
  text = json.dumps(collect_report(), indent=2, ensure_ascii=False) + "\n"
20
  sys.stdout.write(text)
21
  try:
 
12
 
13
 
14
  def main(argv: Sequence[str] | None = None) -> None:
15
+ parser = argparse.ArgumentParser(
16
  prog="fcc-doctor",
17
  description="Print and copy FCC diagnostics for a bug report.",
18
+ )
19
+ parser.add_argument(
20
+ "--disconnect-integrations",
21
+ action="store_true",
22
+ help="Disconnect recognized FCC editor/app integrations before uninstalling.",
23
+ )
24
+ args = parser.parse_args(argv)
25
+ if args.disconnect_integrations:
26
+ from free_claude_code.harnesses.integration_cleanup import (
27
+ IntegrationCleanupError,
28
+ disconnect_integrations,
29
+ )
30
+
31
+ try:
32
+ disconnect_integrations()
33
+ except IntegrationCleanupError as exc:
34
+ print(str(exc), file=sys.stderr)
35
+ raise SystemExit(1) from None
36
+ print("Disconnected recognized FCC editor/app integrations.")
37
+ return
38
  text = json.dumps(collect_report(), indent=2, ensure_ascii=False) + "\n"
39
  sys.stdout.write(text)
40
  try:
src/free_claude_code/cli/launchers/agy.py ADDED
@@ -0,0 +1,95 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from __future__ import annotations
2
+ """Installed Antigravity CLI launcher with an explicit native FCC connection."""
3
+
4
+ import re
5
+ import sys
6
+ from collections.abc import Sequence
7
+
8
+ from free_claude_code.config.loader import get_settings
9
+ from free_claude_code.config.server_urls import local_admin_url
10
+ from free_claude_code.harnesses import antigravity_integration
11
+ from free_claude_code.harnesses.environment import client_environment
12
+ from free_claude_code.harnesses.launch import NativeCheck, PreparedLaunch
13
+ from free_claude_code.harnesses.resources import LaunchResources
14
+
15
+ from .catalog_http import fetch_proxy_model_catalog
16
+ from .runner import (
17
+ HarnessSpec,
18
+ LaunchContext,
19
+ LaunchError,
20
+ launch_harness,
21
+ version_at_least,
22
+ )
23
+
24
+ _VERSION_PATTERN = re.compile(r"(?m)^\s*(\d+)\.(\d+)\.(\d+)(?:\+[0-9A-Za-z.-]+)?\s*$")
25
+ _INSTALL_HINT = (
26
+ "Install Antigravity CLI with `irm https://antigravity.google/cli/install.ps1 | iex` "
27
+ "on Windows or `curl -fsSL https://antigravity.google/cli/install.sh | bash` "
28
+ "on macOS/Linux. Upgrade an existing installation with: agy update"
29
+ )
30
+
31
+
32
+ def _configure(
33
+ ctx: LaunchContext, args: list[str], _files: LaunchResources
34
+ ) -> PreparedLaunch:
35
+ admin_url = ctx.proxy_root_url.rstrip("/") + "/admin/integrations"
36
+ path = antigravity_integration.config_path()
37
+ state = antigravity_integration.state_path()
38
+ try:
39
+ if not antigravity_integration.status(path, state)["connected"]:
40
+ raise antigravity_integration.DisconnectedError()
41
+ catalog = fetch_proxy_model_catalog(ctx.proxy_root_url, ctx.auth_token)
42
+ antigravity_integration.configure(path, state, catalog)
43
+ except antigravity_integration.DisconnectedError as exc:
44
+ raise LaunchError(
45
+ f"{exc}\nConnect it on the Integrations page first:\n{admin_url}"
46
+ ) from None
47
+ if args and not args[0].startswith("-"):
48
+ # Native subcommands dispatch before the session flag parser.
49
+ command = [ctx.binary_path, *args]
50
+ else:
51
+ command = [ctx.binary_path, "--model", catalog.default_model_id, *args]
52
+ return PreparedLaunch(
53
+ command,
54
+ client_environment(
55
+ ctx.base_env,
56
+ proxy_root_url=ctx.proxy_root_url,
57
+ updates={
58
+ "GEMINI_API_KEY": ctx.auth_token,
59
+ "GOOGLE_GEMINI_BASE_URL": ctx.proxy_root_url.rstrip("/"),
60
+ },
61
+ ),
62
+ )
63
+
64
+
65
+ SPEC = HarnessSpec(
66
+ binary_name="agy",
67
+ display_name="Antigravity CLI",
68
+ install_hint=_INSTALL_HINT,
69
+ configure=_configure,
70
+ compatibility_check=NativeCheck(
71
+ ("--version",),
72
+ lambda output: version_at_least(output, _VERSION_PATTERN, (1, 4, 7)),
73
+ "FCC requires Antigravity CLI 1.4.7 or newer.",
74
+ ),
75
+ )
76
+
77
+
78
+ def launch(argv: Sequence[str] | None = None) -> None:
79
+ args = list(sys.argv[1:] if argv is None else argv)
80
+ if args in (["--connect"], ["--disconnect"]):
81
+ try:
82
+ url = local_admin_url(get_settings()) + "/integrations"
83
+ except (OSError, ValueError):
84
+ print(
85
+ "Could not load FCC settings. Run fcc-server to check the configuration.",
86
+ file=sys.stderr,
87
+ )
88
+ else:
89
+ print(
90
+ "Manage the Antigravity CLI connection on the Integrations page:\n"
91
+ f"{url}\nStart FCC with fcc-server if it is not running.",
92
+ file=sys.stderr,
93
+ )
94
+ raise SystemExit(1)
95
+ launch_harness(SPEC, args)
src/free_claude_code/cli/launchers/opencode_config.py CHANGED
@@ -40,6 +40,9 @@ def build_opencode_config(
40
  "settings": {
41
  "baseURL": proxy_v1_url(proxy_root_url),
42
  "apiKey": f"{{env:{OPENCODE_API_KEY_ENV}}}",
 
 
 
43
  },
44
  }
45
  default_model = f"{OPENCODE_PROVIDER_ID}/{default_model_id}"
 
40
  "settings": {
41
  "baseURL": proxy_v1_url(proxy_root_url),
42
  "apiKey": f"{{env:{OPENCODE_API_KEY_ENV}}}",
43
+ "headerTimeout": False,
44
+ "chunkTimeout": False,
45
+ "timeout": False,
46
  },
47
  }
48
  default_model = f"{OPENCODE_PROVIDER_ID}/{default_model_id}"
src/free_claude_code/cli/macos_app.py ADDED
@@ -0,0 +1,148 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from __future__ import annotations
2
+ """Construct the macOS application around the installed managed Python runtime."""
3
+
4
+ import os
5
+ import platform
6
+ import plistlib
7
+ import shutil
8
+ import subprocess
9
+ import sys
10
+ import sysconfig
11
+ import tempfile
12
+ from importlib.metadata import metadata
13
+ from importlib.resources import files
14
+ from pathlib import Path
15
+
16
+ from packaging.specifiers import SpecifierSet
17
+
18
+ from free_claude_code.cli.desktop_assets import export_app_icon
19
+
20
+ BUNDLE_ID = "io.github.alishahryar1.free-claude-code"
21
+ OWNER_FILE = ".free-claude-code-owner"
22
+
23
+
24
+ def runtime_settings() -> tuple[dict[str, str], str]:
25
+ """Keep the venv executable while locating the standalone runtime's library."""
26
+ if (
27
+ platform.python_implementation() != "CPython"
28
+ or sysconfig.get_config_var("Py_GIL_DISABLED")
29
+ or sysconfig.get_config_var("Py_DEBUG")
30
+ ):
31
+ raise ValueError(
32
+ "The desktop app requires a GIL-enabled CPython release build."
33
+ )
34
+ version = platform.python_version()
35
+ requirement = metadata("free-claude-code")["Requires-Python"]
36
+ if not requirement or version not in SpecifierSet(requirement):
37
+ raise ValueError(
38
+ "The installed Python version does not match Free Claude Code."
39
+ )
40
+ architecture = platform.machine()
41
+ if architecture not in {"arm64", "x86_64"}:
42
+ raise ValueError(f"Unsupported macOS Python architecture: {architecture}")
43
+ # Do not resolve this symlink: CPython needs its venv path for pyvenv.cfg.
44
+ executable = Path(sys.executable).absolute()
45
+ if not executable.is_file() or not os.access(executable, os.X_OK):
46
+ raise ValueError("The installed Python executable is missing or inaccessible.")
47
+ library_name = sysconfig.get_config_var("LDLIBRARY")
48
+ if not isinstance(library_name, str) or Path(library_name).name != library_name:
49
+ raise ValueError("The managed Python shared library is unavailable.")
50
+ library = Path(sys.base_prefix) / "lib" / library_name
51
+ if not library.is_file():
52
+ raise ValueError(f"The managed Python shared library is missing: {library}")
53
+ return {
54
+ "PythonExecutable": str(executable),
55
+ "PythonLibrary": str(library.absolute()),
56
+ "PythonVersion": version,
57
+ }, architecture
58
+
59
+
60
+ def is_owned(app: Path) -> bool:
61
+ if app.is_symlink() or not app.is_dir():
62
+ return False
63
+ # Existing shell bundles used Contents directly. Signed bundles use Resources.
64
+ for directory in (app / "Contents/Resources", app / "Contents"):
65
+ marker = directory / OWNER_FILE
66
+ if marker.is_file() and marker.read_text(encoding="utf-8").strip() == BUNDLE_ID:
67
+ return True
68
+ return False
69
+
70
+
71
+ def install() -> Path:
72
+ """Sign a complete sibling bundle before replacing an FCC-owned application."""
73
+ if sys.platform != "darwin":
74
+ raise ValueError("Application installation is supported only on macOS.")
75
+ home = Path.home()
76
+ app = home / "Applications/Free Claude Code.app"
77
+ if (app.exists() or app.is_symlink()) and not is_owned(app):
78
+ raise ValueError(
79
+ f"An app not managed by Free Claude Code already exists at {app}."
80
+ )
81
+ settings, architecture = runtime_settings()
82
+ template = (
83
+ files("free_claude_code")
84
+ .joinpath("assets", "macos", f"fcc-desktop-{architecture}")
85
+ .read_bytes()
86
+ )
87
+ app.parent.mkdir(parents=True, exist_ok=True)
88
+ root = Path(tempfile.mkdtemp(prefix=".fcc-app-", dir=app.parent))
89
+ backup = root / "previous.app"
90
+ published = False
91
+ try:
92
+ staged = root / app.name
93
+ contents = staged / "Contents"
94
+ resources = contents / "Resources"
95
+ resources.mkdir(parents=True)
96
+ (contents / "MacOS").mkdir()
97
+ executable = contents / "MacOS/fcc-desktop"
98
+ executable.write_bytes(template)
99
+ executable.chmod(0o755)
100
+ export_app_icon(resources / "AppIcon.icns")
101
+ (resources / OWNER_FILE).write_text(BUNDLE_ID + "\n", encoding="utf-8")
102
+ with (resources / "runtime.plist").open("wb") as stream:
103
+ plistlib.dump(settings, stream)
104
+ with (contents / "Info.plist").open("wb") as stream:
105
+ plistlib.dump(
106
+ {
107
+ "CFBundleDisplayName": "Free Claude Code",
108
+ "CFBundleExecutable": "fcc-desktop",
109
+ "CFBundleIdentifier": BUNDLE_ID,
110
+ "CFBundleIconFile": "AppIcon",
111
+ "CFBundleName": "Free Claude Code",
112
+ "CFBundlePackageType": "APPL",
113
+ "LSMultipleInstancesProhibited": True,
114
+ "LSUIElement": True,
115
+ },
116
+ stream,
117
+ )
118
+ subprocess.run(
119
+ ["/usr/bin/codesign", "--force", "--sign", "-", str(staged)], check=True
120
+ )
121
+ subprocess.run(
122
+ ["/usr/bin/codesign", "--verify", "--strict", str(staged)], check=True
123
+ )
124
+ if app.exists():
125
+ app.rename(backup)
126
+ try:
127
+ staged.rename(app)
128
+ except OSError:
129
+ if backup.exists():
130
+ backup.rename(app)
131
+ raise
132
+ published = True
133
+ finally:
134
+ if published or not backup.exists():
135
+ shutil.rmtree(root)
136
+ else:
137
+ print(
138
+ f"The previous application is preserved at {backup}.", file=sys.stderr
139
+ )
140
+ desktop = home / "Desktop/Free Claude Code.app"
141
+ desktop.parent.mkdir(parents=True, exist_ok=True)
142
+ if desktop.is_symlink() and desktop.readlink() == app:
143
+ return app
144
+ if desktop.exists() or desktop.is_symlink():
145
+ print(f"A non-FCC item already exists at {desktop}; leaving it unchanged.")
146
+ else:
147
+ desktop.symlink_to(app, target_is_directory=True)
148
+ return app
src/free_claude_code/cli/uvicorn_server.py CHANGED
@@ -1,18 +1,46 @@
1
  from __future__ import annotations
2
  """Uvicorn adapter imported only by the server worker."""
3
 
 
4
  import socket
5
  from collections.abc import Awaitable, Callable
6
  from copy import deepcopy
 
7
 
8
  import uvicorn
9
 
10
  from free_claude_code.config.logging_config import InterceptHandler
11
 
12
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
13
  def uvicorn_log_config(*, console: bool) -> dict[str, object]:
14
  """Send Uvicorn records to FCC's file and optional standard console handlers."""
15
  config = deepcopy(uvicorn.config.LOGGING_CONFIG)
 
 
16
  if not console:
17
  config["formatters"] = {}
18
  config["handlers"] = {}
 
1
  from __future__ import annotations
2
  """Uvicorn adapter imported only by the server worker."""
3
 
4
+ import logging
5
  import socket
6
  from collections.abc import Awaitable, Callable
7
  from copy import deepcopy
8
+ from urllib.parse import parse_qsl, urlencode
9
 
10
  import uvicorn
11
 
12
  from free_claude_code.config.logging_config import InterceptHandler
13
 
14
 
15
+ class QueryCredentialFilter(logging.Filter):
16
+ """Mask query API keys before either Uvicorn output handler sees them."""
17
+
18
+ def filter(self, record: logging.LogRecord) -> bool:
19
+ if (
20
+ record.name == "uvicorn.access"
21
+ and isinstance(record.args, tuple)
22
+ and len(record.args) == 5
23
+ ):
24
+ address, method, target, version, status = record.args
25
+ if isinstance(target, str) and "?" in target:
26
+ path, _, query = target.partition("?")
27
+ pairs = parse_qsl(query, keep_blank_values=True)
28
+ if any(key == "key" for key, _ in pairs):
29
+ query = urlencode(
30
+ [
31
+ (key, "[REDACTED]" if key == "key" else value)
32
+ for key, value in pairs
33
+ ]
34
+ )
35
+ record.args = (address, method, f"{path}?{query}", version, status)
36
+ return True
37
+
38
+
39
  def uvicorn_log_config(*, console: bool) -> dict[str, object]:
40
  """Send Uvicorn records to FCC's file and optional standard console handlers."""
41
  config = deepcopy(uvicorn.config.LOGGING_CONFIG)
42
+ config["filters"] = {"query_credentials": {"()": QueryCredentialFilter}}
43
+ config["loggers"]["uvicorn.access"]["filters"] = ["query_credentials"]
44
  if not console:
45
  config["formatters"] = {}
46
  config["handlers"] = {}
src/free_claude_code/config/admin/manifest.py CHANGED
@@ -388,46 +388,6 @@ _NON_PROVIDER_FIELDS: tuple[ConfigFieldSpec, ...] = (
388
  settings_attr="whisper_model",
389
  session_sensitive=True,
390
  ),
391
- ConfigFieldSpec(
392
- "FAST_PREFIX_DETECTION",
393
- "Fast Prefix Detection",
394
- "runtime",
395
- "boolean",
396
- settings_attr="fast_prefix_detection",
397
- advanced=True,
398
- ),
399
- ConfigFieldSpec(
400
- "ENABLE_NETWORK_PROBE_MOCK",
401
- "Network Probe Mock",
402
- "runtime",
403
- "boolean",
404
- settings_attr="enable_network_probe_mock",
405
- advanced=True,
406
- ),
407
- ConfigFieldSpec(
408
- "ENABLE_TITLE_GENERATION_SKIP",
409
- "Title Generation Skip",
410
- "runtime",
411
- "boolean",
412
- settings_attr="enable_title_generation_skip",
413
- advanced=True,
414
- ),
415
- ConfigFieldSpec(
416
- "ENABLE_SUGGESTION_MODE_SKIP",
417
- "Suggestion Mode Skip",
418
- "runtime",
419
- "boolean",
420
- settings_attr="enable_suggestion_mode_skip",
421
- advanced=True,
422
- ),
423
- ConfigFieldSpec(
424
- "ENABLE_FILEPATH_EXTRACTION_MOCK",
425
- "Filepath Extraction Mock",
426
- "runtime",
427
- "boolean",
428
- settings_attr="enable_filepath_extraction_mock",
429
- advanced=True,
430
- ),
431
  ConfigFieldSpec(
432
  "ENABLE_WEB_SERVER_TOOLS",
433
  "Web Server Tools",
 
388
  settings_attr="whisper_model",
389
  session_sensitive=True,
390
  ),
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
391
  ConfigFieldSpec(
392
  "ENABLE_WEB_SERVER_TOOLS",
393
  "Web Server Tools",
src/free_claude_code/config/admin/provider_manifest.py CHANGED
@@ -20,7 +20,7 @@ class ProviderFieldOverride(TypedDict, total=False):
20
 
21
  _PROVIDER_FIELD_OVERRIDES: dict[str, ProviderFieldOverride] = {
22
  "ANTHROPIC_API_KEY": {
23
- "description": "Anthropic API access with separate usage billing. Messages requests use Anthropic tools and client thinking controls. Native fallback uses other Anthropic models. FCC local web and prompt optimizations apply to compatibility routes.",
24
  },
25
  "OPENAI_API_KEY": {
26
  "label": "OpenAI API Keys",
@@ -28,7 +28,8 @@ _PROVIDER_FIELD_OVERRIDES: dict[str, ProviderFieldOverride] = {
28
  },
29
  "OPENAI_PROXY": {
30
  "description": (
31
- "Optional proxy used for OpenAI sign-in and ChatGPT Codex requests. "
 
32
  "Changing it restarts FCC."
33
  ),
34
  "restart_required": True,
@@ -475,6 +476,7 @@ def _proxy_field_specs() -> tuple[ConfigFieldSpec, ...]:
475
  ConfigFieldSpec(
476
  key=_settings_env_key(descriptor.proxy_attr),
477
  label=f"{descriptor.display_name} Proxy",
 
478
  section_id="providers",
479
  field_type="secret",
480
  settings_attr=descriptor.proxy_attr,
 
20
 
21
  _PROVIDER_FIELD_OVERRIDES: dict[str, ProviderFieldOverride] = {
22
  "ANTHROPIC_API_KEY": {
23
+ "description": "Anthropic API access with separate usage billing. Messages requests use Anthropic tools and client thinking controls. Native fallback uses other Anthropic models. FCC local web tools apply to compatibility routes.",
24
  },
25
  "OPENAI_API_KEY": {
26
  "label": "OpenAI API Keys",
 
28
  },
29
  "OPENAI_PROXY": {
30
  "description": (
31
+ "Optional network proxy. Leave blank unless you use one. "
32
+ "Used for OpenAI sign-in and ChatGPT Codex requests. "
33
  "Changing it restarts FCC."
34
  ),
35
  "restart_required": True,
 
476
  ConfigFieldSpec(
477
  key=_settings_env_key(descriptor.proxy_attr),
478
  label=f"{descriptor.display_name} Proxy",
479
+ description="Optional network proxy. Leave blank unless you use one",
480
  section_id="providers",
481
  field_type="secret",
482
  settings_attr=descriptor.proxy_attr,
src/free_claude_code/config/constants.py CHANGED
@@ -1,7 +1,7 @@
1
  from __future__ import annotations
2
  """Shared defaults used by config models and provider adapters."""
3
 
4
- DEFAULT_MODEL = "nvidia_nim/nvidia/nemotron-3-super-120b-a12b"
5
 
6
  # Client context allocation when the provider reports no model limit.
7
  DEFAULT_MODEL_CONTEXT_TOKENS = 200_000
 
1
  from __future__ import annotations
2
  """Shared defaults used by config models and provider adapters."""
3
 
4
+ DEFAULT_MODEL = "nvidia_nim/nvidia/nemotron-3-ultra-550b-a55b"
5
 
6
  # Client context allocation when the provider reports no model limit.
7
  DEFAULT_MODEL_CONTEXT_TOKENS = 200_000
src/free_claude_code/config/env_migrations.py CHANGED
@@ -141,8 +141,6 @@ def settings_env_keys() -> frozenset[str]:
141
 
142
  keys: set[str] = set()
143
  for name, field in Settings.model_fields.items():
144
- if name == "nim":
145
- continue
146
  alias = field.validation_alias
147
  if not isinstance(alias, str):
148
  raise AssertionError(f"Settings field {name!r} needs one string alias")
@@ -312,7 +310,7 @@ def _log_model_repairs(before: Mapping[str, str], after: Mapping[str, str]) -> N
312
  changed = sorted(key for key in before if before[key] != after.get(key))
313
  if changed:
314
  logger.info(
315
- "Repaired retired provider selections in managed config: {}",
316
  ", ".join(changed),
317
  )
318
 
 
141
 
142
  keys: set[str] = set()
143
  for name, field in Settings.model_fields.items():
 
 
144
  alias = field.validation_alias
145
  if not isinstance(alias, str):
146
  raise AssertionError(f"Settings field {name!r} needs one string alias")
 
310
  changed = sorted(key for key in before if before[key] != after.get(key))
311
  if changed:
312
  logger.info(
313
+ "Repaired retired model selections in managed config: {}",
314
  ", ".join(changed),
315
  )
316
 
src/free_claude_code/config/loader.py CHANGED
@@ -2,7 +2,7 @@ from __future__ import annotations
2
  """Canonical managed-config loading, precedence, provenance, and caching."""
3
 
4
  import os
5
- from collections.abc import Iterator, Mapping
6
  from contextlib import contextmanager
7
  from dataclasses import dataclass
8
  from dataclasses import field as dataclass_field
@@ -66,7 +66,7 @@ class ManagedConfigStore:
66
  self._lock_path = config_lock_path()
67
 
68
  @contextmanager
69
- def _storage_lock(self) -> Iterator[None]:
70
  lock = InterprocessFileLock(self._lock_path)
71
  if not lock.acquire(wait=True, timeout=10.0):
72
  raise TimeoutError(
@@ -208,8 +208,6 @@ def clear_settings_cache() -> None:
208
  def _settings_aliases() -> dict[str, str]:
209
  aliases: dict[str, str] = {}
210
  for name, field in Settings.model_fields.items():
211
- if name == "nim":
212
- continue
213
  alias = field.validation_alias
214
  if not isinstance(alias, str):
215
  raise AssertionError(f"Settings field {name!r} needs one string alias")
 
2
  """Canonical managed-config loading, precedence, provenance, and caching."""
3
 
4
  import os
5
+ from collections.abc import Generator, Mapping
6
  from contextlib import contextmanager
7
  from dataclasses import dataclass
8
  from dataclasses import field as dataclass_field
 
66
  self._lock_path = config_lock_path()
67
 
68
  @contextmanager
69
+ def _storage_lock(self) -> Generator[None]:
70
  lock = InterprocessFileLock(self._lock_path)
71
  if not lock.acquire(wait=True, timeout=10.0):
72
  raise TimeoutError(
 
208
  def _settings_aliases() -> dict[str, str]:
209
  aliases: dict[str, str] = {}
210
  for name, field in Settings.model_fields.items():
 
 
211
  alias = field.validation_alias
212
  if not isinstance(alias, str):
213
  raise AssertionError(f"Settings field {name!r} needs one string alias")
src/free_claude_code/config/model_refs.py CHANGED
@@ -22,6 +22,13 @@ from .reasoning import ReasoningPreference
22
  RETIRED_PROVIDER_IDS = frozenset({"github_models"})
23
 
24
 
 
 
 
 
 
 
 
25
  class ModelFallback(BaseModel):
26
  """One ordered fallback's model, reasoning preference, and key number."""
27
 
@@ -98,6 +105,8 @@ def normalize_retired_model_settings(
98
  normalized[key] = DEFAULT_MODEL if key == "MODEL" else ""
99
  else:
100
  normalized.pop(key)
 
 
101
  try:
102
  fallbacks = decode_model_fallbacks(normalized.get("MODEL_FALLBACKS"))
103
  except ValueError:
@@ -106,6 +115,18 @@ def normalize_retired_model_settings(
106
  retained = tuple(
107
  entry for entry in fallbacks if not is_retired_model_ref(entry.model)
108
  )
 
 
 
 
 
 
 
 
 
 
 
 
109
  if retained:
110
  normalized["MODEL_FALLBACKS"] = encode_model_fallbacks(retained)
111
  elif preserve_empty_overrides:
@@ -125,12 +146,23 @@ class ConfiguredChatModelRef:
125
 
126
 
127
  class ChatModelConfig(Protocol):
128
- model: str
129
- model_fable: str | None
130
- model_opus: str | None
131
- model_sonnet: str | None
132
- model_haiku: str | None
133
- model_fallbacks: tuple[ModelFallback, ...] | None
 
 
 
 
 
 
 
 
 
 
 
134
 
135
 
136
  def split_provider_model_ref(model_ref: str) -> tuple[str, str]:
 
22
  RETIRED_PROVIDER_IDS = frozenset({"github_models"})
23
 
24
 
25
+ def replace_retired_model_ref(model_ref: str) -> str:
26
+ """Replace the exact retired hosted endpoint, retaining independent providers."""
27
+ if model_ref.strip() == "nvidia_nim/nvidia/nemotron-3-super-120b-a12b":
28
+ return "nvidia_nim/nvidia/nemotron-3-ultra-550b-a55b"
29
+ return model_ref
30
+
31
+
32
  class ModelFallback(BaseModel):
33
  """One ordered fallback's model, reasoning preference, and key number."""
34
 
 
105
  normalized[key] = DEFAULT_MODEL if key == "MODEL" else ""
106
  else:
107
  normalized.pop(key)
108
+ elif key in normalized:
109
+ normalized[key] = replace_retired_model_ref(normalized[key])
110
  try:
111
  fallbacks = decode_model_fallbacks(normalized.get("MODEL_FALLBACKS"))
112
  except ValueError:
 
115
  retained = tuple(
116
  entry for entry in fallbacks if not is_retired_model_ref(entry.model)
117
  )
118
+ original_pairs = [(entry.model, entry.api_key_number) for entry in retained]
119
+ if len(original_pairs) != len(set(original_pairs)):
120
+ return normalized # Settings must still reject pre-existing duplicates.
121
+ migrated: list[ModelFallback] = []
122
+ seen: set[tuple[str, int]] = set()
123
+ for entry in retained:
124
+ model = replace_retired_model_ref(entry.model)
125
+ pair = (model, entry.api_key_number)
126
+ if pair not in seen:
127
+ seen.add(pair)
128
+ migrated.append(entry.model_copy(update={"model": model}))
129
+ retained = tuple(migrated)
130
  if retained:
131
  normalized["MODEL_FALLBACKS"] = encode_model_fallbacks(retained)
132
  elif preserve_empty_overrides:
 
146
 
147
 
148
  class ChatModelConfig(Protocol):
149
+ @property
150
+ def model(self) -> str: ...
151
+
152
+ @property
153
+ def model_fable(self) -> str | None: ...
154
+
155
+ @property
156
+ def model_opus(self) -> str | None: ...
157
+
158
+ @property
159
+ def model_sonnet(self) -> str | None: ...
160
+
161
+ @property
162
+ def model_haiku(self) -> str | None: ...
163
+
164
+ @property
165
+ def model_fallbacks(self) -> tuple[ModelFallback, ...] | None: ...
166
 
167
 
168
  def split_provider_model_ref(model_ref: str) -> tuple[str, str]:
src/free_claude_code/config/nim.py DELETED
@@ -1,122 +0,0 @@
1
- from __future__ import annotations
2
- """NVIDIA NIM settings (fixed values, no env config)."""
3
-
4
- from pydantic import BaseModel, ConfigDict, Field, ValidationInfo, field_validator
5
-
6
- from free_claude_code.config.constants import ANTHROPIC_DEFAULT_MAX_OUTPUT_TOKENS
7
-
8
-
9
- class NimSettings(BaseModel):
10
- """Fixed NVIDIA NIM settings (not configurable via env)."""
11
-
12
- temperature: float = Field(
13
- 1.0, ge=0.0, le=2.0, description="Sampling temperature, must be >=0 and <=2."
14
- )
15
- top_p: float = Field(
16
- 0.95,
17
- ge=0.95,
18
- le=0.95,
19
- description="FCC's fixed NVIDIA NIM nucleus sampling probability.",
20
- )
21
- top_k: int = -1
22
- max_tokens: int = Field(
23
- ANTHROPIC_DEFAULT_MAX_OUTPUT_TOKENS,
24
- ge=1,
25
- description="Maximum number of tokens in output.",
26
- )
27
- presence_penalty: float = Field(0.0, ge=-2.0, le=2.0)
28
- frequency_penalty: float = Field(0.0, ge=-2.0, le=2.0)
29
- min_p: float = Field(
30
- 0.0, ge=0.0, le=1.0, description="Minimum probability threshold [0,1]."
31
- )
32
- repetition_penalty: float = Field(
33
- 1.0, ge=0.0, description="Penalty for repeated tokens. Must be >=0."
34
- )
35
- seed: int | None = None
36
- stop: str | None = None
37
- parallel_tool_calls: bool = True
38
- ignore_eos: bool = False
39
- min_tokens: int = Field(0, ge=0, description="Minimum tokens in the response.")
40
- chat_template: str | None = None
41
- request_id: str | None = None
42
-
43
- model_config = ConfigDict(extra="forbid")
44
-
45
- @field_validator("top_k", mode="before")
46
- @classmethod
47
- def validate_top_k(cls, v, info: ValidationInfo):
48
- if v is None or v == "":
49
- return -1
50
- int_v = int(v)
51
- if int_v < -1:
52
- raise ValueError(f"{info.field_name} must be -1 or >= 0")
53
- return int_v
54
-
55
- @field_validator(
56
- "temperature",
57
- "top_p",
58
- "min_p",
59
- "presence_penalty",
60
- "frequency_penalty",
61
- "repetition_penalty",
62
- mode="before",
63
- )
64
- @classmethod
65
- def validate_float_fields(cls, v, info: ValidationInfo):
66
- field_defaults = {
67
- "temperature": 1.0,
68
- "top_p": 0.95,
69
- "min_p": 0.0,
70
- "presence_penalty": 0.0,
71
- "frequency_penalty": 0.0,
72
- "repetition_penalty": 1.0,
73
- }
74
- if v is None or v == "":
75
- key = info.field_name or "temperature"
76
- return field_defaults.get(key, 1.0)
77
- try:
78
- val = float(v)
79
- except (TypeError, ValueError) as err:
80
- raise ValueError(
81
- f"{info.field_name} must be a float. Got {type(v).__name__}."
82
- ) from err
83
- return val
84
-
85
- @field_validator("max_tokens", "min_tokens", mode="before")
86
- @classmethod
87
- def validate_int_fields(cls, v, info: ValidationInfo):
88
- field_defaults = {
89
- "max_tokens": ANTHROPIC_DEFAULT_MAX_OUTPUT_TOKENS,
90
- "min_tokens": 0,
91
- }
92
- if v is None or v == "":
93
- key = info.field_name or "max_tokens"
94
- return field_defaults.get(key, ANTHROPIC_DEFAULT_MAX_OUTPUT_TOKENS)
95
- try:
96
- val = int(v)
97
- except (TypeError, ValueError) as err:
98
- raise ValueError(
99
- f"{info.field_name} must be an int. Got {type(v).__name__}."
100
- ) from err
101
- return val
102
-
103
- @field_validator("seed", mode="before")
104
- @classmethod
105
- def parse_optional_int(cls, v, info: ValidationInfo):
106
- if v == "" or v is None:
107
- return None
108
- try:
109
- return int(v)
110
- except (TypeError, ValueError) as err:
111
- raise ValueError(
112
- f"{info.field_name} must be an int or empty/None."
113
- ) from err
114
-
115
- @field_validator("stop", "chat_template", "request_id", mode="before")
116
- @classmethod
117
- def parse_optional_str(cls, v, info: ValidationInfo):
118
- if v == "":
119
- return None
120
- if v is not None and not isinstance(v, str):
121
- return str(v)
122
- return v
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
src/free_claude_code/config/settings.py CHANGED
@@ -17,7 +17,6 @@ from .api_keys import ApiKeyList
17
  from .constants import DEFAULT_MODEL, HTTP_CONNECT_TIMEOUT_DEFAULT
18
  from .custom_providers import CustomProviderDefinition, decode_custom_providers
19
  from .model_refs import ModelFallback, ModelFallbacks
20
- from .nim import NimSettings
21
  from .provider_catalog import (
22
  BEDROCK_DEFAULT_BASE,
23
  PROVIDER_CATALOG,
@@ -59,6 +58,7 @@ class Settings(BaseModel):
59
  """Validated application settings with no file or process I/O."""
60
 
61
  model_config = ConfigDict(
 
62
  validate_default=True,
63
  populate_by_name=True,
64
  extra="ignore",
@@ -651,30 +651,6 @@ class Settings(BaseModel):
651
  validation_alias="HTTP_CONNECT_TIMEOUT",
652
  )
653
 
654
- # ==================== Fast Prefix Detection ====================
655
- fast_prefix_detection: bool = Field(
656
- default=True,
657
- validation_alias="FAST_PREFIX_DETECTION",
658
- )
659
-
660
- # ==================== Optimizations ====================
661
- enable_network_probe_mock: bool = Field(
662
- default=True,
663
- validation_alias="ENABLE_NETWORK_PROBE_MOCK",
664
- )
665
- enable_title_generation_skip: bool = Field(
666
- default=True,
667
- validation_alias="ENABLE_TITLE_GENERATION_SKIP",
668
- )
669
- enable_suggestion_mode_skip: bool = Field(
670
- default=True,
671
- validation_alias="ENABLE_SUGGESTION_MODE_SKIP",
672
- )
673
- enable_filepath_extraction_mock: bool = Field(
674
- default=True,
675
- validation_alias="ENABLE_FILEPATH_EXTRACTION_MOCK",
676
- )
677
-
678
  # ==================== Local web server tools (web_search / web_fetch) ====================
679
  # On by default to match Claude Code's normal web-tool availability.
680
  enable_web_server_tools: bool = Field(
@@ -722,9 +698,6 @@ class Settings(BaseModel):
722
  default=False, validation_alias="DEBUG_SUBAGENT_STACK"
723
  )
724
 
725
- # ==================== NIM Settings ====================
726
- nim: NimSettings = Field(default_factory=NimSettings)
727
-
728
  # ==================== Voice Note Transcription ====================
729
  voice_note_enabled: bool = Field(
730
  default=True, validation_alias="VOICE_NOTE_ENABLED"
 
17
  from .constants import DEFAULT_MODEL, HTTP_CONNECT_TIMEOUT_DEFAULT
18
  from .custom_providers import CustomProviderDefinition, decode_custom_providers
19
  from .model_refs import ModelFallback, ModelFallbacks
 
20
  from .provider_catalog import (
21
  BEDROCK_DEFAULT_BASE,
22
  PROVIDER_CATALOG,
 
58
  """Validated application settings with no file or process I/O."""
59
 
60
  model_config = ConfigDict(
61
+ frozen=True,
62
  validate_default=True,
63
  populate_by_name=True,
64
  extra="ignore",
 
651
  validation_alias="HTTP_CONNECT_TIMEOUT",
652
  )
653
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
654
  # ==================== Local web server tools (web_search / web_fetch) ====================
655
  # On by default to match Claude Code's normal web-tool availability.
656
  enable_web_server_tools: bool = Field(
 
698
  default=False, validation_alias="DEBUG_SUBAGENT_STACK"
699
  )
700
 
 
 
 
701
  # ==================== Voice Note Transcription ====================
702
  voice_note_enabled: bool = Field(
703
  default=True, validation_alias="VOICE_NOTE_ENABLED"
src/free_claude_code/core/anthropic/tokens.py CHANGED
@@ -7,7 +7,10 @@ from typing import Any
7
 
8
  from loguru import logger
9
 
10
- from free_claude_code.core.token_estimation import estimate_text_tokens
 
 
 
11
 
12
  from .content import get_block_attr, get_block_type
13
  from .models import Message, NativeTokenCountMessage, SystemContent, Tool
@@ -28,8 +31,8 @@ def _image_tokens(block: Any) -> int:
28
  if isinstance(source, dict):
29
  data = source.get("data") or source.get("base64") or ""
30
  if isinstance(data, str) and data:
31
- return max(85, len(data) // 3000)
32
- return 765
33
 
34
 
35
  def get_token_count(
 
7
 
8
  from loguru import logger
9
 
10
+ from free_claude_code.core.token_estimation import (
11
+ estimate_image_tokens,
12
+ estimate_text_tokens,
13
+ )
14
 
15
  from .content import get_block_attr, get_block_type
16
  from .models import Message, NativeTokenCountMessage, SystemContent, Tool
 
31
  if isinstance(source, dict):
32
  data = source.get("data") or source.get("base64") or ""
33
  if isinstance(data, str) and data:
34
+ return estimate_image_tokens(len(data))
35
+ return estimate_image_tokens()
36
 
37
 
38
  def get_token_count(
src/free_claude_code/core/async_rwlock.py CHANGED
@@ -2,7 +2,7 @@ from __future__ import annotations
2
  """Concurrent readers and exclusive writers with preference for queued writes."""
3
 
4
  import asyncio
5
- from collections.abc import AsyncIterator
6
  from contextlib import asynccontextmanager
7
 
8
  from .async_tasks import wait_owned
@@ -16,7 +16,7 @@ class AsyncReadWriteLock:
16
  self._waiting_writers = 0
17
 
18
  @asynccontextmanager
19
- async def read(self) -> AsyncIterator[None]:
20
  async with self._condition:
21
  await self._condition.wait_for(
22
  lambda: not self._writer and not self._waiting_writers
@@ -28,7 +28,7 @@ class AsyncReadWriteLock:
28
  await wait_owned(asyncio.create_task(self._release(writer=False)))
29
 
30
  @asynccontextmanager
31
- async def write(self) -> AsyncIterator[None]:
32
  async with self._condition:
33
  self._waiting_writers += 1
34
  try:
 
2
  """Concurrent readers and exclusive writers with preference for queued writes."""
3
 
4
  import asyncio
5
+ from collections.abc import AsyncGenerator
6
  from contextlib import asynccontextmanager
7
 
8
  from .async_tasks import wait_owned
 
16
  self._waiting_writers = 0
17
 
18
  @asynccontextmanager
19
+ async def read(self) -> AsyncGenerator[None]:
20
  async with self._condition:
21
  await self._condition.wait_for(
22
  lambda: not self._writer and not self._waiting_writers
 
28
  await wait_owned(asyncio.create_task(self._release(writer=False)))
29
 
30
  @asynccontextmanager
31
+ async def write(self) -> AsyncGenerator[None]:
32
  async with self._condition:
33
  self._waiting_writers += 1
34
  try:
src/free_claude_code/core/continuation_overlap.py CHANGED
@@ -8,20 +8,11 @@ from typing import Any
8
  try:
9
  import simplejson
10
  except ImportError:
11
- import json
12
- class _SimpleJsonFallback:
13
- @staticmethod
14
- def dumps(obj, *args, **kwargs):
15
- kwargs.pop('use_decimal', None)
16
- return json.dumps(obj, *args, **kwargs)
17
- @staticmethod
18
- def loads(s, *args, **kwargs):
19
- kwargs.pop('use_decimal', None)
20
- return json.loads(s, *args, **kwargs)
21
- simplejson = _SimpleJsonFallback()
22
 
23
  from .delivered_response import RECOVERY_RETENTION_BYTES
24
  from .failures import ExecutionFailure, FailureKind
 
25
 
26
  type PartKey = tuple[int | str, int]
27
 
@@ -88,7 +79,7 @@ class ContinuationOverlap:
88
 
89
  def _feed(self, data: dict[str, Any]) -> list[dict[str, Any]]:
90
  kind = data.get("type", "")
91
- if kind in {"error", "response.error", "response.failed"}:
92
  self._frames.clear()
93
  self._frame_bytes = 0
94
  return [data]
@@ -96,13 +87,18 @@ class ContinuationOverlap:
96
  return self._render(_Frame(data, self._texts(data)))
97
  texts = self._texts(data)
98
  frame = _Frame(data, texts)
99
- terminal = kind in {
100
- "response.completed",
101
- "response.incomplete",
102
- "message_stop",
103
- } or (
104
- kind == "message_delta"
105
- and data.get("delta", {}).get("stop_reason") is not None
 
 
 
 
 
106
  )
107
  if not self._pending:
108
  if kind in {"response.completed", "response.incomplete"}:
@@ -168,6 +164,12 @@ class ContinuationOverlap:
168
  return identity
169
 
170
  def _texts(self, data: dict[str, Any]) -> list[_Text]:
 
 
 
 
 
 
171
  kind = data.get("type", "")
172
  index = data.get("index", 0)
173
  if kind == "content_block_start":
@@ -314,6 +316,17 @@ class ContinuationOverlap:
314
  and all(text.delta for text in frame.texts)
315
  and not any(text.value.get(text.field) for text in frame.texts)
316
  ):
 
 
 
 
 
 
 
 
 
 
 
317
  return []
318
  return [frame.data]
319
 
 
8
  try:
9
  import simplejson
10
  except ImportError:
11
+ import json as simplejson
 
 
 
 
 
 
 
 
 
 
12
 
13
  from .delivered_response import RECOVERY_RETENTION_BYTES
14
  from .failures import ExecutionFailure, FailureKind
15
+ from .gemini.response import gemini_finished, gemini_parts
16
 
17
  type PartKey = tuple[int | str, int]
18
 
 
79
 
80
  def _feed(self, data: dict[str, Any]) -> list[dict[str, Any]]:
81
  kind = data.get("type", "")
82
+ if kind in {"error", "response.error", "response.failed"} or "error" in data:
83
  self._frames.clear()
84
  self._frame_bytes = 0
85
  return [data]
 
87
  return self._render(_Frame(data, self._texts(data)))
88
  texts = self._texts(data)
89
  frame = _Frame(data, texts)
90
+ terminal = (
91
+ gemini_finished(data)
92
+ or kind
93
+ in {
94
+ "response.completed",
95
+ "response.incomplete",
96
+ "message_stop",
97
+ }
98
+ or (
99
+ kind == "message_delta"
100
+ and data.get("delta", {}).get("stop_reason") is not None
101
+ )
102
  )
103
  if not self._pending:
104
  if kind in {"response.completed", "response.incomplete"}:
 
164
  return identity
165
 
166
  def _texts(self, data: dict[str, Any]) -> list[_Text]:
167
+ if "candidates" in data:
168
+ return [
169
+ _Text(("gemini", 0), (0, 0), part, "text", delta=True)
170
+ for part in gemini_parts(data)
171
+ if "text" in part and not part.get("thought")
172
+ ]
173
  kind = data.get("type", "")
174
  index = data.get("index", 0)
175
  if kind == "content_block_start":
 
316
  and all(text.delta for text in frame.texts)
317
  and not any(text.value.get(text.field) for text in frame.texts)
318
  ):
319
+ if "candidates" in frame.data:
320
+ for candidate in frame.data["candidates"]:
321
+ content = candidate.get("content", {})
322
+ content["parts"] = [
323
+ part
324
+ for part in content.get("parts", [])
325
+ if part.get("text")
326
+ or any(key not in {"text", "thought"} for key in part)
327
+ ]
328
+ if gemini_finished(frame.data) or gemini_parts(frame.data):
329
+ return [frame.data]
330
  return []
331
  return [frame.data]
332
 
src/free_claude_code/core/continuation_stream.py CHANGED
@@ -12,6 +12,7 @@ from .delivered_response import (
12
  responses_text_constraint,
13
  )
14
  from .failures import ExecutionFailure, FailureKind
 
15
 
16
 
17
  class ContinuationStream:
@@ -90,6 +91,11 @@ class ContinuationStream:
90
  return [self._map(event) for event in events]
91
 
92
  def _map(self, data: dict[str, Any]) -> dict[str, Any]:
 
 
 
 
 
93
  if not self.handoff:
94
  if isinstance(data.get("index"), int):
95
  data["index"] += self._offset
@@ -117,6 +123,17 @@ class ContinuationStream:
117
  """Finish the response only after preceding frames have been published."""
118
  if not self.active:
119
  return data
 
 
 
 
 
 
 
 
 
 
 
120
  kind = data.get("type")
121
  response = data.get("response")
122
  output = response.get("output", []) if isinstance(response, dict) else []
 
12
  responses_text_constraint,
13
  )
14
  from .failures import ExecutionFailure, FailureKind
15
+ from .gemini.response import gemini_finished
16
 
17
 
18
  class ContinuationStream:
 
91
  return [self._map(event) for event in events]
92
 
93
  def _map(self, data: dict[str, Any]) -> dict[str, Any]:
94
+ if self.delivered.wire_api == "gemini":
95
+ for key in ("responseId", "modelVersion"):
96
+ if key in self.delivered.header:
97
+ data[key] = self.delivered.header[key]
98
+ return data
99
  if not self.handoff:
100
  if isinstance(data.get("index"), int):
101
  data["index"] += self._offset
 
123
  """Finish the response only after preceding frames have been published."""
124
  if not self.active:
125
  return data
126
+ if self.delivered.wire_api == "gemini":
127
+ if gemini_finished(data) and not self.handoff and not self.made_progress:
128
+ raise ExecutionFailure(
129
+ kind=FailureKind.UPSTREAM,
130
+ status_code=502,
131
+ message="The provider continuation completed without adding output.",
132
+ retryable=False,
133
+ )
134
+ # Per-attempt usage cannot describe a synthesized multi-attempt answer.
135
+ data.pop("usageMetadata", None)
136
+ return data
137
  kind = data.get("type")
138
  response = data.get("response")
139
  output = response.get("output", []) if isinstance(response, dict) else []
src/free_claude_code/core/delivered_response.py CHANGED
@@ -7,28 +7,17 @@ from typing import Any, Literal
7
  try:
8
  import simplejson
9
  except ImportError:
10
- import json
11
- class _SimpleJsonFallback:
12
- @staticmethod
13
- def dumps(obj, *args, **kwargs):
14
- kwargs.pop('use_decimal', None)
15
- return json.dumps(obj, *args, **kwargs)
16
- @staticmethod
17
- def loads(s, *args, **kwargs):
18
- kwargs.pop('use_decimal', None)
19
- return json.loads(s, *args, **kwargs)
20
- simplejson = _SimpleJsonFallback()
21
 
 
22
  from .history_replay import (
23
  readable_reasoning,
24
  unencrypted_responses_replay,
25
  )
26
  from .openai_responses import is_client_search
27
- from .stream_recovery import DeliveredPrefix
28
  from .token_estimation import estimate_text_tokens
29
 
30
- RECOVERY_RETENTION_BYTES = 1_048_576
31
-
32
  CLIENT_CALL_TYPES = frozenset(
33
  {
34
  "function_call",
@@ -82,7 +71,7 @@ def responses_text_constraint(data: dict[str, Any]) -> str | None:
82
  class DeliveredResponse:
83
  """Track this response only; the harness continues to own conversation history."""
84
 
85
- def __init__(self, wire_api: Literal["messages", "responses"]) -> None:
86
  self.wire_api = wire_api
87
  self.items: dict[int, dict[str, Any]] = {}
88
  self.closed: set[int] = set()
@@ -121,6 +110,9 @@ class DeliveredResponse:
121
  return self._limited
122
 
123
  def observe(self, data: dict[str, Any], *, count_progress: bool = True) -> None:
 
 
 
124
  kind = data.get("type", "")
125
  if not isinstance(kind, str):
126
  self.unsafe_reason = "unknown_event"
@@ -169,6 +161,44 @@ class DeliveredResponse:
169
  if self._limited:
170
  self._limited_output_tokens += estimate_text_tokens(text)
171
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
172
  def _new_snapshot_text(self, data: dict[str, Any]) -> bool:
173
  """Cumulative text is progress only when it adds published characters."""
174
  kind = data.get("type")
@@ -401,7 +431,12 @@ class DeliveredResponse:
401
  has_calls = False
402
  for index, item in sorted(self.items.items()):
403
  kind = item.get("type")
404
- if self.wire_api == "messages":
 
 
 
 
 
405
  if kind == "text":
406
  text.append(item.get("text", ""))
407
  elif kind == "thinking":
@@ -435,6 +470,7 @@ class DeliveredResponse:
435
  unsafe_reason is None,
436
  self.can_handoff,
437
  unsafe_reason,
 
438
  )
439
 
440
  def output_tokens(self, output: list[dict[str, Any]] | None = None) -> int:
@@ -463,6 +499,8 @@ class DeliveredResponse:
463
  )
464
 
465
  def closing_events(self) -> list[dict[str, Any]]:
 
 
466
  events: list[dict[str, Any]] = []
467
  for index, item in sorted(self.items.items()):
468
  if index in self.closed:
@@ -518,6 +556,13 @@ class DeliveredResponse:
518
 
519
  def handoff_events(self) -> list[dict[str, Any]]:
520
  """Signal handoff; the public writer fills output and usage after closure."""
 
 
 
 
 
 
 
521
  if self.wire_api == "messages":
522
  return [
523
  {
 
7
  try:
8
  import simplejson
9
  except ImportError:
10
+ import json as simplejson
 
 
 
 
 
 
 
 
 
 
11
 
12
+ from .gemini.response import gemini_parts
13
  from .history_replay import (
14
  readable_reasoning,
15
  unencrypted_responses_replay,
16
  )
17
  from .openai_responses import is_client_search
18
+ from .stream_recovery import RECOVERY_RETENTION_BYTES, DeliveredPrefix
19
  from .token_estimation import estimate_text_tokens
20
 
 
 
21
  CLIENT_CALL_TYPES = frozenset(
22
  {
23
  "function_call",
 
71
  class DeliveredResponse:
72
  """Track this response only; the harness continues to own conversation history."""
73
 
74
+ def __init__(self, wire_api: Literal["messages", "responses", "gemini"]) -> None:
75
  self.wire_api = wire_api
76
  self.items: dict[int, dict[str, Any]] = {}
77
  self.closed: set[int] = set()
 
110
  return self._limited
111
 
112
  def observe(self, data: dict[str, Any], *, count_progress: bool = True) -> None:
113
+ if self.wire_api == "gemini":
114
+ self._gemini(data, count_progress=count_progress)
115
+ return
116
  kind = data.get("type", "")
117
  if not isinstance(kind, str):
118
  self.unsafe_reason = "unknown_event"
 
161
  if self._limited:
162
  self._limited_output_tokens += estimate_text_tokens(text)
163
 
164
+ def _gemini(self, data: dict[str, Any], *, count_progress: bool) -> None:
165
+ if not self.header:
166
+ self.header = {
167
+ key: data[key] for key in ("responseId", "modelVersion") if key in data
168
+ }
169
+ parts = gemini_parts(data)
170
+ if count_progress and any(
171
+ part.get("text") or part.get("functionCall") for part in parts
172
+ ):
173
+ self.progress += 1
174
+ for part in parts:
175
+ if not self._retain(part):
176
+ self._limited_output_tokens += estimate_text_tokens(
177
+ part.get("text", "")
178
+ )
179
+ continue
180
+ if part.get("thoughtSignature"):
181
+ self.unsafe_reason = "opaque_reasoning"
182
+ if any(
183
+ key not in {"text", "thought", "thoughtSignature", "functionCall"}
184
+ for key in part
185
+ ):
186
+ self.unsafe_reason = "unsupported_output"
187
+ index = max(self.items, default=-1) + 1
188
+ previous = self.items.get(index - 1)
189
+ if (
190
+ previous is not None
191
+ and "text" in part
192
+ and "text" in previous
193
+ and set(part) <= {"text", "thought"}
194
+ and set(previous) <= {"text", "thought"}
195
+ and part.get("thought", False) == previous.get("thought", False)
196
+ ):
197
+ previous["text"] += part["text"]
198
+ else:
199
+ self.items[index] = deepcopy(part)
200
+ self.closed.add(index)
201
+
202
  def _new_snapshot_text(self, data: dict[str, Any]) -> bool:
203
  """Cumulative text is progress only when it adds published characters."""
204
  kind = data.get("type")
 
431
  has_calls = False
432
  for index, item in sorted(self.items.items()):
433
  kind = item.get("type")
434
+ if self.wire_api == "gemini":
435
+ if "text" in item:
436
+ (thinking if item.get("thought") else text).append(item["text"])
437
+ if "functionCall" in item:
438
+ has_calls = True
439
+ elif self.wire_api == "messages":
440
  if kind == "text":
441
  text.append(item.get("text", ""))
442
  elif kind == "thinking":
 
470
  unsafe_reason is None,
471
  self.can_handoff,
472
  unsafe_reason,
473
+ retained_bytes=self._bytes,
474
  )
475
 
476
  def output_tokens(self, output: list[dict[str, Any]] | None = None) -> int:
 
499
  )
500
 
501
  def closing_events(self) -> list[dict[str, Any]]:
502
+ if self.wire_api == "gemini":
503
+ return []
504
  events: list[dict[str, Any]] = []
505
  for index, item in sorted(self.items.items()):
506
  if index in self.closed:
 
556
 
557
  def handoff_events(self) -> list[dict[str, Any]]:
558
  """Signal handoff; the public writer fills output and usage after closure."""
559
+ if self.wire_api == "gemini":
560
+ return [
561
+ {
562
+ **deepcopy(self.header),
563
+ "candidates": [{"index": 0, "finishReason": "STOP"}],
564
+ }
565
+ ]
566
  if self.wire_api == "messages":
567
  return [
568
  {