patch_models.py

"""Patch backend: replace AVAILABLE_MODELS with OpenRouter models + routing."""

import ast
import os
import re

AGENT_FILE = "/app/backend/routes/agent.py"
LLM_PARAMS_FILE = "/app/agent/core/llm_params.py"

# === Step 1: Patch model_ids.py ===
IDS_FILE = "/app/agent/core/model_ids.py"
with open(IDS_FILE) as f:
    content = f.read()

# Swap out old model definitions for new ones
content = content.replace(
    "CLAUDE_OPUS_48_MODEL_ID", "__DELETED_CLAUDE"
)
content = content.replace(
    'KIMI_K27_CODE_MODEL_ID = "moonshotai/Kimi-K2.7-Code:novita"',
    'KIMI_K27_CODE_MODEL_ID = "deepseek-ai/DeepSeek-V4-Pro"',
)
content = content.replace(
    'MINIMAX_M3_MODEL_ID = "MiniMaxAI/MiniMax-M3:novita"',
    'MINIMAX_M3_MODEL_ID = "deepseek-ai/DeepSeek-V4-Flash"',
)
content = content.replace(
    'GLM_52_MODEL_ID = "zai-org/GLM-5.2:novita"',
    'GLM_52_MODEL_ID = "openai/deepseek/deepseek-v4-flash"',
)
content = content.replace(
    'DEEPSEEK_V4_PRO_MODEL_ID = "deepseek-ai/DeepSeek-V4-Pro:novita"',
    'DEEPSEEK_V4_PRO_MODEL_ID = "nvidia/nemotron-3-super-120b-a12b:free"',
)

# Add new model IDs
new_ids = """
GEMMA_4_31B_FREE_MODEL_ID = "openai/google/gemma-4-31b-it:free"
TENCENT_HY3_FREE_MODEL_ID = "openai/tencent/hy3:free"
LLAMA_3_3_70B_FREE_MODEL_ID = "openai/meta-llama/llama-3.3-70b-instruct:free"
LLAMA_3_1_8B_MODEL_ID = "openai/meta-llama/llama-3.1-8b-instruct"
LAGUNA_M1_FREE_MODEL_ID = "openai/poolside/laguna-m.1:free"
LAGUNA_S21_FREE_MODEL_ID = "openai/poolside/laguna-s-2.1:free"
NEX_N2_MINI_MODEL_ID = "openai/nex-agi/nex-n2-mini"
LING_3_0_FLASH_FREE_MODEL_ID = "openai/inclusionai/ling-3.0-flash:free"
"""
if "GEMMA_4_31B_FREE_MODEL_ID" not in content:
    content = content.replace(
        "DEEPSEEK_V4_PRO_MODEL_ID",
        new_ids + "\nDEEPSEEK_V4_PRO_MODEL_ID"
    )

# Update HOSTED_MODEL_IDS
hosted_start = content.find("HOSTED_MODEL_IDS = {")
if hosted_start >= 0:
    hosted_end = content.find("}", hosted_start) + 1
    hosted_block = content[hosted_start:hosted_end]
    new_hosted = "HOSTED_MODEL_IDS = {\n    GPT_55_MODEL_ID,\n    KIMI_K27_CODE_MODEL_ID,\n    MINIMAX_M3_MODEL_ID,\n    GLM_52_MODEL_ID,\n    DEEPSEEK_V4_PRO_MODEL_ID,\n    GEMMA_4_31B_FREE_MODEL_ID,\n    TENCENT_HY3_FREE_MODEL_ID,\n    LLAMA_3_3_70B_FREE_MODEL_ID,\n    LLAMA_3_1_8B_MODEL_ID,\n    LAGUNA_M1_FREE_MODEL_ID,\n    LAGUNA_S21_FREE_MODEL_ID,\n    NEX_N2_MINI_MODEL_ID,\n    LING_3_0_FLASH_FREE_MODEL_ID,\n}"
    content = content.replace(hosted_block, new_hosted)

with open(IDS_FILE, "w") as f:
    f.write(content)
print("OK: model_ids.py patched")

# === Step 2: Patch agent.py ===
with open(AGENT_FILE) as f:
    content = f.read()

content = content.replace(
    "DEFAULT_MODEL_ID = GLM_52_MODEL_ID",
    'DEFAULT_MODEL_ID = "openai/tencent/hy3:free"'
)
content = content.replace(
    "DEFAULT_GPT_MODEL_ID = GPT_55_MODEL_ID",
    "DEFAULT_GPT_MODEL_ID = LAGUNA_S21_FREE_MODEL_ID"
)

# Replace import block
old_import = (
    "from agent.core.model_ids import (\n"
    "    CLAUDE_OPUS_48_MODEL_ID,\n"
    "    DEEPSEEK_V4_PRO_MODEL_ID,\n"
    "    GLM_52_MODEL_ID,\n"
    "    GPT_55_MODEL_ID,\n"
    "    KIMI_K27_CODE_MODEL_ID,\n"
    "    MINIMAX_M3_MODEL_ID,\n"
    "    strip_huggingface_model_prefix,\n"
    ")"
)
new_import = (
    "from agent.core.model_ids import (\n"
    "    DEEPSEEK_V4_PRO_MODEL_ID,\n"
    "    GLM_52_MODEL_ID,\n"
    "    GPT_55_MODEL_ID,\n"
    "    KIMI_K27_CODE_MODEL_ID,\n"
    "    MINIMAX_M3_MODEL_ID,\n"
    "    GEMMA_4_31B_FREE_MODEL_ID,\n"
    "    TENCENT_HY3_FREE_MODEL_ID,\n"
    "    LLAMA_3_3_70B_FREE_MODEL_ID,\n"
    "    LLAMA_3_1_8B_MODEL_ID,\n"
    "    LAGUNA_M1_FREE_MODEL_ID,\n"
    "    LAGUNA_S21_FREE_MODEL_ID,\n"
    "    NEX_N2_MINI_MODEL_ID,\n"
    "    LING_3_0_FLASH_FREE_MODEL_ID,\n"
    "    strip_huggingface_model_prefix,\n"
    ")"
)
content = content.replace(old_import, new_import)

# Replace _available_models function
new_func = (
    "def _available_models() -> list[dict[str, Any]]:\n"
    "    models = [\n"
    '        {"id": TENCENT_HY3_FREE_MODEL_ID, "label": "Tencent HY3:free", "recommended": True},\n'
    '        {"id": GEMMA_4_31B_FREE_MODEL_ID, "label": "Gemma 4 31B:free", "recommended": True},\n'
    '        {"id": LLAMA_3_3_70B_FREE_MODEL_ID, "label": "Llama 3.3 70B:free", "recommended": True},\n'
    '        {"id": LAGUNA_M1_FREE_MODEL_ID, "label": "Laguna M.1:free", "recommended": True},\n'
    '        {"id": DEFAULT_GPT_MODEL_ID, "label": "Laguna S 2.1:free", "recommended": True},\n'
    '        {"id": NEX_N2_MINI_MODEL_ID, "label": "nex-agi/nex-n2-mini", "recommended": True},\n'
    '        {"id": LING_3_0_FLASH_FREE_MODEL_ID, "label": "inclusionai/ling-3.0-flash:free", "recommended": True},\n'
    '        {"id": LLAMA_3_1_8B_MODEL_ID, "label": "Llama 3.1 8B"},\n'
    '        {"id": KIMI_K27_CODE_MODEL_ID, "label": "DeepSeek V4 Pro"},\n'
    '        {"id": MINIMAX_M3_MODEL_ID, "label": "DeepSeek V4 Flash"},\n'
    '        {"id": DEFAULT_MODEL_ID, "label": "DeepSeek V4 Flash"},\n'
    '        {"id": DEEPSEEK_V4_PRO_MODEL_ID, "label": "Nemotron 3 Super 120B"},\n'
    "    ]\n"
    "    return models"
)

# Find the old function by pattern
func_match = re.search(
    r"def _available_models\(\)\s*->\s*list\[dict\[str,\s*Any\]\]:.*?return models",
    content, re.DOTALL
)
if func_match:
    content = content[:func_match.start()] + new_func + content[func_match.end():]
    print("OK: Replaced _available_models()")
else:
    # Fallback: try exact match
    old_func = (
        "def _available_models() -> list[dict[str, Any]]:\n"
        "    models = [\n"
        '        {"id": CLAUDE_OPUS_48_MODEL_ID, "label": "Claude Opus 4.8"},\n'
        '        {"id": DEFAULT_GPT_MODEL_ID, "label": "GPT-5.5"},\n'
        '        {"id": KIMI_K27_CODE_MODEL_ID, "label": "Kimi K2.7 Code"},\n'
        '        {"id": MINIMAX_M3_MODEL_ID, "label": "MiniMax M3"},\n'
        '        {"id": DEFAULT_MODEL_ID, "label": "GLM 5.2", "recommended": True},\n'
        '        {"id": DEEPSEEK_V4_PRO_MODEL_ID, "label": "DeepSeek V4 Pro"},\n'
        "    ]\n"
        "    return models"
    )
    if old_func in content:
        content = content.replace(old_func, new_func)
        print("OK: String replace of _available_models()")
    else:
        print("FAIL: Cannot find _available_models()")
        print("Content excerpt around 'def _available_models':")
        idx = content.find("def _available_models")
        if idx >= 0:
            print(content[idx:idx+500])
        import sys; sys.exit(1)

# Update title gen model
content = content.replace(
    '"openai/gpt-oss-120b:cerebras",',
    '"huggingface/deepseek-ai/DeepSeek-V4-Pro",'
)

# Validate syntax
try:
    ast.parse(content)
    print("OK: agent.py syntax OK")
except SyntaxError as e:
    print(f"FAIL: agent.py syntax error: {e}")
    raise

with open(AGENT_FILE, "w") as f:
    f.write(content)
print("OK: agent.py patched")

# === Step 3: Patch _resolve_llm_params ===
with open(LLM_PARAMS_FILE) as f:
    llm_content = f.read()

if "normalized_model.startswith" not in llm_content:
    api_key_find = "api_key = _resolve_hf_router_token(session_hf_token)"
    if api_key_find in llm_content:
        line_end = llm_content.find("\n", llm_content.find(api_key_find) + len(api_key_find)) + 1
        routing_insert = (
            '    # Route openai/-prefixed models to OpenRouter\n'
            '    if normalized_model.startswith("openai/"):\n'
            '        return {\n'
            '            "model": normalized_model,\n'
            '            "api_base": "https://openrouter.ai/api/v1",\n'
            '            "api_key": os.environ.get("OPENROUTER_API_KEY") or api_key or "",\n'
            '        }\n\n'
        )
        llm_content = llm_content[:line_end] + routing_insert + llm_content[line_end:]
        print("OK: Patched _resolve_llm_params")
    else:
        print("WARN: Could not find api_key line")
else:
    print("OK: llm_params.py already patched")

try:
    ast.parse(llm_content)
    print("OK: llm_params.py syntax OK")
except SyntaxError as e:
    print(f"FAIL: llm_params.py syntax error: {e}")
    raise

with open(LLM_PARAMS_FILE, "w") as f:
    f.write(llm_content)

print("OK: ALL BACKEND PATCHES APPLIED")