patch_models.py

"""Patch backend: replace AVAILABLE_MODELS (remove old 5, add 16 new) + OpenRouter routing."""

import ast
import os

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

# === Step 1: Replace _available_models() to return ONLY new models ===
with open(AGENT_FILE) as f:
    content = f.read()

# Replace the default model constants first
content = content.replace(
    'DEFAULT_FREE_MODEL_ID = KIMI_K26_MODEL_ID',
    'DEFAULT_FREE_MODEL_ID = "openai/openrouter/owl-alpha"'
)

# Replace _available_models() function body to return only new models
old_func = '''def _available_models() -> list[dict[str, Any]]:
    models = [
        {
            "id": DEFAULT_OPUS_MODEL_ID,
            "label": "Claude Opus 4.8",
            "provider": "huggingface",
            "recommended": True,
        },
        {
            "id": DEFAULT_GPT_MODEL_ID,
            "label": "GPT-5.5",
            "provider": "huggingface",
        },
        {
            "id": DEFAULT_FREE_MODEL_ID,
            "label": "Kimi K2.6",
            "provider": "huggingface",
        },
        {
            "id": MINIMAX_M27_MODEL_ID,
            "label": "MiniMax M2.7",
            "provider": "huggingface",
        },
        {
            "id": GLM_51_MODEL_ID,
            "label": "GLM 5.1",
            "provider": "huggingface",
        },
        {
            "id": DEEPSEEK_V4_PRO_MODEL_ID,
            "label": "DeepSeek V4 Pro",
            "provider": "huggingface",
        },
    ]
    return models'''

new_func = '''def _available_models() -> list[dict[str, Any]]:
    models = [
        {
            "id": "openai/openrouter/owl-alpha",
            "label": "Owl Alpha",
            "provider": "openrouter",
            "tier": "free",
            "recommended": True,
        },
        {
            "id": "deepseek-ai/DeepSeek-V4-Pro",
            "label": "DeepSeek V4 Pro",
            "provider": "huggingface",
            "tier": "free",
            "recommended": True,
        },
        {
            "id": "deepseek-ai/DeepSeek-V4-Flash",
            "label": "DeepSeek V4 Flash",
            "provider": "huggingface",
            "tier": "free",
        },
        {
            "id": "openai/gpt-oss-120b",
            "label": "gpt-oss-120b",
            "provider": "openrouter",
            "tier": "free",
        },
        {
            "id": "Qwen/Qwen3-Coder-Next",
            "label": "Qwen3 Coder Next",
            "provider": "huggingface",
            "tier": "free",
            "recommended": True,
        },
        {
            "id": "google/gemma-3-1b-it",
            "label": "Gemma 3 1B",
            "provider": "huggingface",
            "tier": "free",
            "recommended": True,
        },
        {
            "id": "openai/google/gemini-2.0-flash-001",
            "label": "Gemini 2.0 Flash",
            "provider": "openrouter",
            "tier": "free",
        },
        {
            "id": "openai/deepseek/deepseek-v4-flash",
            "label": "DeepSeek V4 Flash (OR)",
            "provider": "openrouter",
            "tier": "paid",
        },
        {
            "id": "openai/google/gemma-4-9b-it:free",
            "label": "Gemma 4 9B",
            "provider": "openrouter",
            "tier": "free",
            "recommended": True,
        },
        {
            "id": "openai/google/gemma-4-31b-it:free",
            "label": "Gemma 4 31B",
            "provider": "openrouter",
            "tier": "free",
            "recommended": True,
        },
        {
            "id": "openai/tencent/hy3-preview",
            "label": "Hy3 Preview",
            "provider": "openrouter",
            "tier": "paid",
            "recommended": True,
        },
        {
            "id": "openai/cohere/north-mini-code:free",
            "label": "north-mini-code:free",
            "provider": "openrouter",
            "tier": "free",
            "recommended": True,
        },
        {
            "id": "openai/nvidia/nemotron-3-super-120b-a12b:free",
            "label": "Nemotron 3 Super 120B",
            "provider": "openrouter",
            "tier": "free",
        },
        {
            "id": "openai/nvidia/nemotron-3-ultra-550b-a55b:free",
            "label": "Nemotron 3 Ultra 550B",
            "provider": "openrouter",
            "tier": "free",
            "recommended": True,
        },
        {
            "id": "openai/poolside/laguna-m.1:free",
            "label": "Laguna M.1",
            "provider": "openrouter",
            "tier": "free",
            "recommended": True,
        },
        {
            "id": "openai/poolside/laguna-xs.2:free",
            "label": "Laguna XS.2",
            "provider": "openrouter",
            "tier": "free",
        },
    ]
    return models'''

if old_func in content:
    content = content.replace(old_func, new_func)
    print("OK: Replaced _available_models() with 16 new models")
else:
    print("WARN: Could not find old _available_models(), trying extend approach...")
    target = "AVAILABLE_MODELS = _available_models()\n"
    if target in content:
        content = content.replace(
            target,
            target + "AVAILABLE_MODELS.clear()\n" +
            "AVAILABLE_MODELS.extend([\n"
            '    {"id": "openai/openrouter/owl-alpha", "label": "Owl Alpha", "provider": "openrouter", "tier": "free", "recommended": True},\n'
            '    {"id": "deepseek-ai/DeepSeek-V4-Pro", "label": "DeepSeek V4 Pro", "provider": "huggingface", "tier": "free", "recommended": True},\n'
            '    {"id": "deepseek-ai/DeepSeek-V4-Flash", "label": "DeepSeek V4 Flash", "provider": "huggingface", "tier": "free"},\n'
            '    {"id": "openai/gpt-oss-120b", "label": "gpt-oss-120b", "provider": "openrouter", "tier": "free"},\n'
            '    {"id": "Qwen/Qwen3-Coder-Next", "label": "Qwen3 Coder Next", "provider": "huggingface", "tier": "free", "recommended": True},\n'
            '    {"id": "google/gemma-3-1b-it", "label": "Gemma 3 1B", "provider": "huggingface", "tier": "free", "recommended": True},\n'
            '    {"id": "openai/google/gemini-2.0-flash-001", "label": "Gemini 2.0 Flash", "provider": "openrouter", "tier": "free"},\n'
            '    {"id": "openai/deepseek/deepseek-v4-flash", "label": "DeepSeek V4 Flash (OR)", "provider": "openrouter", "tier": "paid"},\n'
            '    {"id": "openai/google/gemma-4-9b-it:free", "label": "Gemma 4 9B", "provider": "openrouter", "tier": "free", "recommended": True},\n'
            '    {"id": "openai/google/gemma-4-31b-it:free", "label": "Gemma 4 31B", "provider": "openrouter", "tier": "free", "recommended": True},\n'
            '    {"id": "openai/tencent/hy3-preview", "label": "Hy3 Preview", "provider": "openrouter", "tier": "paid", "recommended": True},\n'
            '    {"id": "openai/cohere/north-mini-code:free", "label": "north-mini-code:free", "provider": "openrouter", "tier": "free", "recommended": True},\n'
            '    {"id": "openai/nvidia/nemotron-3-super-120b-a12b:free", "label": "Nemotron 3 Super 120B", "provider": "openrouter", "tier": "free"},\n'
            '    {"id": "openai/nvidia/nemotron-3-ultra-550b-a55b:free", "label": "Nemotron 3 Ultra 550B", "provider": "openrouter", "tier": "free", "recommended": True},\n'
            '    {"id": "openai/poolside/laguna-m.1:free", "label": "Laguna M.1", "provider": "openrouter", "tier": "free", "recommended": True},\n'
            '    {"id": "openai/poolside/laguna-xs.2:free", "label": "Laguna XS.2", "provider": "openrouter", "tier": "free"},\n'
            "])\n"
            'DEFAULT_FREE_MODEL_ID = "openai/openrouter/owl-alpha"\n'
        )
        print("OK: Used extend approach")
    else:
        print("FAIL: Cannot patch agent.py!")
        import sys
        sys.exit(1)

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

with open(AGENT_FILE, "w") as f:
    f.write(content)

# === Step 2: Patch _resolve_llm_params for OpenRouter routing ===
# IMPORTANT: Keep the "openai/" prefix when sending to OpenRouter!
# OpenRouter accepts the full slug like "openai/openrouter/owl-alpha".
# LiteLLM needs "openai/" for provider detection.
with open(LLM_PARAMS_FILE) as f:
    llm_content = f.read()

# First, try to replace the full block (original version)
old_block = """    hf_model = normalized_model
    api_key = _resolve_hf_router_token(session_hf_token)
    params = {
        "model": f"openai/{hf_model}",
        "api_base": HF_ROUTER_BASE_URL,
        "api_key": api_key,
    }"""

new_block = """    hf_model = normalized_model
    api_key = _resolve_hf_router_token(session_hf_token)

    # === PATCH: Route ALL openai/-prefixed models to OpenRouter ===
    # Keep the "openai/" prefix — both LiteLLM and OpenRouter need it.
    if normalized_model.startswith("openai/"):
        return {
            "model": normalized_model,
            "api_base": "https://openrouter.ai/api/v1",
            "api_key": os.environ.get("OPENROUTER_API_KEY") or api_key or "",
        }
    # === END PATCH ===

    params = {
        "model": f"openai/{hf_model}",
        "api_base": HF_ROUTER_BASE_URL,
        "api_key": api_key,
    }"""

if old_block in llm_content:
    llm_content = llm_content.replace(old_block, new_block)
    print("OK: Patched _resolve_llm_params (catch-all openai/ -> OR, keep prefix)")
elif "normalized_model[len" in llm_content or "actual_model" in llm_content:
    # Revert the strip approach — keep openai/ prefix
    idx = llm_content.find("if normalized_model.startswith(\"openai/\"):")
    if idx >= 0:
        or_start = llm_content.find("# === PATCH:", idx)
        if or_start < 0:
            or_start = idx
        or_end = llm_content.find("# === END PATCH", or_start)
        if or_end < 0:
            or_end = llm_content.find("params = {", or_start)
        
        # Find the return block and fix it
        ret_line = llm_content.find("return {", or_start, or_end)
        if ret_line > 0:
            model_line_start = llm_content.find('"model":', ret_line)
            if model_line_start > 0:
                line_start = llm_content.rfind('\n', 0, model_line_start) + 1
                line_end = llm_content.find('\n', line_start)
                old_model_line = llm_content[line_start:line_end]
                new_model_line = '            "model": normalized_model,'
                llm_content = llm_content.replace(old_model_line, new_model_line)
                print("OK: Reverted to keep openai/ prefix for OpenRouter")
            else:
                print("OK: Could not find model line")
        else:
            print("OK: Could not find return block")
    else:
        print("OK: Already using correct routing")
else:
    print("OK: Routing already correct")

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

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

print("OK: Backend patched - 16 models, ALL openai/ -> OR (keep prefix), GPT-5.5→gpt-oss-120b, Riverflow→north-mini-code:free")