Update app.py

app.py

@@ -62,20 +62,50 @@ MAX_RETRIES      = 2
 # "disabled" → force thinking OFF
 THINKING_MODE = os.getenv("THINKING_MODE", "enabled").strip().lower()
 
-UPSTREAM_HEADERS = {
-    "Content-Type":    "application/json",
+# opencode CLI version to impersonate — update if the CLI releases a new version
+_OPENCODE_VERSION = os.getenv("OPENCODE_VERSION", "1.3.15")
+
+# Static headers — same for every request
+_STATIC_HEADERS = {
+    "Content-Type":        "application/json",
     # identity encoding → no gzip overhead on the hot SSE path
-    "Accept-Encoding": "identity",
-    "Accept":          "application/json",
-    "Origin":          "https://opencode.ai",
-    "Referer":         "https://opencode.ai/",
-    "User-Agent": (
-        "Mozilla/5.0 (Windows NT 10.0; Win64; x64) "
-        "AppleWebKit/537.36 (KHTML, like Gecko) "
-        "Chrome/124.0 Safari/537.36"
-    ),
+    "Accept-Encoding":     "identity",
+    "Accept":              "application/json",
+    "Origin":              "https://opencode.ai",
+    "Referer":             "https://opencode.ai/",
+    # Impersonate the official opencode CLI User-Agent
+    # Without this, Zen treats the client as anonymous → very low rate limits
+    "User-Agent":          f"opencode/latest/{_OPENCODE_VERSION}/cli",
+    # Tell Zen we are the official CLI client
+    "x-opencode-client":   "cli",
 }
 
+# One stable project-ID per process boot (mimics CLI behaviour:
+# same project across requests within a session)
+_PROCESS_PROJECT_ID = uuid.uuid4().hex
+
+
+def _make_upstream_headers(api_key: str) -> dict:
+    """
+    Build a full set of upstream headers for one request.
+
+    The opencode Zen backend rate-limits by client identity.
+    It grants the full 'dailyRequests' quota only to requests
+    that carry the four x-opencode-* headers the official CLI sends.
+    Without them requests fall back to a much smaller 'fallbackValue' limit.
+
+    x-opencode-session and x-opencode-request are re-randomised every
+    call (matching CLI behaviour); x-opencode-project is stable per
+    process boot.
+    """
+    return {
+        **_STATIC_HEADERS,
+        "x-api-key":           api_key,
+        "x-opencode-session":  uuid.uuid4().hex,
+        "x-opencode-project":  _PROCESS_PROJECT_ID,
+        "x-opencode-request":  uuid.uuid4().hex,
+    }
+
 STREAM_RESPONSE_HEADERS = {
     "Cache-Control":    "no-cache",
     "Connection":       "keep-alive",
@@ -89,13 +119,13 @@ STREAM_RESPONSE_HEADERS = {
 
 MODEL_MAP = {
     "claude-sonnet-4-6":         "nemotron-3-super-free",
-    "claude-sonnet-4-5":         "nemotron-3-super-free",
+    "claude-sonnet-4-5":         "deepseek-v4-flash-free",
     "claude-haiku-4-6":          "deepseek-v4-flash-free",
     "claude-haiku-4-5":          "deepseek-v4-flash-free",
-    "claude-haiku-4-5-20251001": "nemotron-3-super-free",
+    "claude-haiku-4-5-20251001": "deepseek-v4-flash-free",
     "claude-opus-4-6":           "nemotron-3-super-free",
 }
-DEFAULT_MODEL = "nemotron-3-super-free"
+DEFAULT_MODEL = "deepseek-v4-flash-free"
 
 _MODELS_CACHE = {
     "object": "list",
@@ -754,7 +784,7 @@ async def messages(
         body, openai_messages, openai_tools,
         openai_tool_choice, mapped_model, stream,
     )
-    headers = {**UPSTREAM_HEADERS, "x-api-key": OPENCODE_API_KEY}
+    headers = _make_upstream_headers(OPENCODE_API_KEY)
 
     logger.info(
         "[%s] %s→%s stream=%s tools=%d thinking=%s msgs=%d",
@@ -778,6 +808,20 @@ async def messages(
                 )
                 response = await client.send(upstream_req, stream=True)
 
+                if response.status_code == 429:
+                    retry_after = float(response.headers.get("retry-after", "5"))
+                    raw_err = (await response.aread()).decode(errors="ignore")
+                    logger.warning(
+                        "[%s] 429 rate-limited — retry-after=%.1fs body=%s",
+                        request_id, retry_after, raw_err[:200]
+                    )
+                    last_error = raw_err
+                    await response.aclose()
+                    await asyncio.sleep(retry_after)
+                    # refresh per-request headers (new random IDs)
+                    headers = _make_upstream_headers(OPENCODE_API_KEY)
+                    continue
+
                 if response.status_code != 200:
                     raw_err = (await response.aread()).decode(errors="ignore")
                     logger.warning("[%s] upstream %d: %s",
@@ -798,6 +842,19 @@ async def messages(
             response = await client.post(
                 UPSTREAM_URL, json=payload, headers=headers
             )
+
+            if response.status_code == 429:
+                retry_after = float(response.headers.get("retry-after", "5"))
+                logger.warning(
+                    "[%s] 429 rate-limited — retry-after=%.1fs",
+                    request_id, retry_after
+                )
+                last_error = response.text
+                await asyncio.sleep(retry_after)
+                # refresh per-request headers (new random IDs)
+                headers = _make_upstream_headers(OPENCODE_API_KEY)
+                continue
+
             if response.status_code != 200:
                 logger.warning("[%s] upstream %d", request_id, response.status_code)
                 last_error = response.text