"""Patch backend: replace AVAILABLE_MODELS (remove old 6, add 25 new) + OpenRouter 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: Replace _available_models() to return ONLY new models === with open(AGENT_FILE) as f: content = f.read() # Make DEFAULT_MODEL_ID point to our default content = content.replace( "DEFAULT_MODEL_ID = GLM_52_MODEL_ID", 'DEFAULT_MODEL_ID = "deepseek-ai/DeepSeek-V4-Pro:novita"' ) content = content.replace( "DEFAULT_GPT_MODEL_ID = GPT_55_MODEL_ID", 'DEFAULT_GPT_MODEL_ID = "openai/gpt-oss-120b"' ) # Find and replace the _available_models() function using regex func_pattern = re.compile( r'def _available_models\(\) -> list\[dict\[str, Any\]\]:\s*\n' r'\s+models\s*=\s*\[.*?\]\s*\n\s+return models', re.DOTALL ) 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:novita", "label": "DeepSeek V4 Pro", "provider": "huggingface", "tier": "free", "recommended": True, }, { "id": "deepseek-ai/DeepSeek-V4-Flash:novita", "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:novita", "label": "Qwen3 Coder Next", "provider": "huggingface", "tier": "free", "recommended": True, }, { "id": "google/gemma-3-1b-it:novita", "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", }, { "id": "openai/qwen/qwen3.7-flash", "label": "Qwen3.7 Flash", "provider": "openrouter", "tier": "free", }, { "id": "openai/inclusionai/ling-2.6-flash", "label": "Ling 2.6 Flash", "provider": "openrouter", "tier": "free", }, { "id": "openai/inclusionai/ling-3.0-flash", "label": "Ling 3.0 Flash", "provider": "openrouter", "tier": "free", }, { "id": "openai/inclusionai/ling-3.0-tiny:free", "label": "Ling 3.0 Tiny", "provider": "openrouter", "tier": "free", }, { "id": "openai/inclusionai/ling-3.0-flash-vl:free", "label": "Ling 3.0 Flash VL", "provider": "openrouter", "tier": "free", "recommended": True, }, { "id": "openai/nex-agi/nex-n2.5-mini:free", "label": "Nex-N2.5-Mini", "provider": "openrouter", "tier": "free", "recommended": True, }, { "id": "openai/nex-agi/nex-n2.5-pro:free", "label": "Nex-N2.5-Pro", "provider": "openrouter", "tier": "free", "recommended": True, }, { "id": "z-ai/glm-5.3-flash", "label": "GLM 5.3 Flash", "provider": "openrouter", "tier": "paid", }, { "id": "deepseek/deepseek-v4-flash-0731", "label": "DeepSeek V4 Flash 0731", "provider": "openrouter", "tier": "free", }, ] return models''' func_match = func_pattern.search(content) if func_match: content = content[:func_match.start()] + new_func + content[func_match.end():] print("OK: Replaced _available_models() with 25 new models via regex") else: print("WARN: Could not find _available_models() via regex, 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:novita", "label": "DeepSeek V4 Pro", "provider": "huggingface", "tier": "free", "recommended": True},\n' ' {"id": "deepseek-ai/DeepSeek-V4-Flash:novita", "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:novita", "label": "Qwen3 Coder Next", "provider": "huggingface", "tier": "free", "recommended": True},\n' ' {"id": "google/gemma-3-1b-it:novita", "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' ' {"id": "openai/qwen/qwen3.7-flash", "label": "Qwen3.7 Flash", "provider": "openrouter", "tier": "free"},\n' ' {"id": "openai/inclusionai/ling-2.6-flash", "label": "Ling 2.6 Flash", "provider": "openrouter", "tier": "free"},\n' ' {"id": "openai/inclusionai/ling-3.0-flash", "label": "Ling 3.0 Flash", "provider": "openrouter", "tier": "free"},\n' ' {"id": "openai/inclusionai/ling-3.0-tiny:free", "label": "Ling 3.0 Tiny", "provider": "openrouter", "tier": "free"},\n' ' {"id": "openai/inclusionai/ling-3.0-flash-vl:free", "label": "Ling 3.0 Flash VL", "provider": "openrouter", "tier": "free", "recommended": True},\n' ' {"id": "openai/nex-agi/nex-n2.5-mini:free", "label": "Nex-N2.5-Mini", "provider": "openrouter", "tier": "free", "recommended": True},\n' ' {"id": "openai/nex-agi/nex-n2.5-pro:free", "label": "Nex-N2.5-Pro", "provider": "openrouter", "tier": "free", "recommended": True},\n' ' {"id": "z-ai/glm-5.3-flash", "label": "GLM 5.3 Flash", "provider": "openrouter", "tier": "paid"},\n' ' {"id": "deepseek/deepseek-v4-flash-0731", "label": "DeepSeek V4 Flash 0731", "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 === with open(LLM_PARAMS_FILE) as f: llm_content = f.read() # Insert the OpenRouter routing before the params block api_key_match = re.search(r'api_key\s*=\s*_resolve_hf_router_token\(session_hf_token\)', llm_content) if api_key_match: line_end = llm_content.find('\n', api_key_match.end()) + 1 if line_end > 0: routing_insert = ''' # === PATCH: Route deepseek/ + z-ai/ + openai/-prefixed models to OpenRouter === # FIX: Add explicit deepseek/ route so "deepseek/deepseek-v4-flash-0731" goes # to OpenRouter (HF Router does NOT have this id -> "model does not exist"). if normalized_model.startswith("deepseek/"): return { "model": normalized_model, "api_base": "https://openrouter.ai/api/v1", "api_key": os.environ.get("OPENROUTER_API_KEY") or api_key or "", } # FIX: Route z-ai/glm-5.3-flash to OpenRouter with the provider prefix so litellm # recognizes the provider. Strip any trailing ":free" suffix. if normalized_model.startswith("z-ai/"): or_model = normalized_model if or_model.endswith(":free"): or_model = or_model[: -len(":free")] return { "model": "openrouter/" + or_model, "api_base": "https://openrouter.ai/api/v1", "api_key": os.environ.get("OPENROUTER_API_KEY") or api_key or "", } # FIX: Strip the "openai/" prefix before sending to OpenRouter. # OpenRouter model ids do NOT include the "openai/" prefix, so sending the # prefixed name causes a 404 (model does not exist) on OpenRouter. Prefix # with "openrouter/" so litellm recognizes the provider. if normalized_model.startswith("openai/"): or_model = "openrouter/" + normalized_model[len("openai/"):] return { "model": or_model, "api_base": "https://openrouter.ai/api/v1", "api_key": os.environ.get("OPENROUTER_API_KEY") or api_key or "", } # === END PATCH === ''' llm_content = llm_content[:line_end] + routing_insert + llm_content[line_end:] print("OK: Patched _resolve_llm_params (deepseek/ + z-ai/[:free stripped] + stripped openai/ -> OR with provider prefix)") else: print("OK: Could not find end of api_key line") else: print("OK: api_key line not found, may already be patched") 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 - 25 models, deepseek/ + z-ai/[:free stripped] + stripped openai/ -> OR") # === Step 3: Patch backend/main.py — safe static cache headers === # FIX: Stale-cache blank page. HF Spaces serves the static build with NO # Cache-Control headers. After every rebuild the hashed asset filename # changes but the browser keeps caching index.html which then references a # bundle hash that no longer exists -> 404 on /assets/index-*.js -> blank # page. Fix: 1yr immutable cache for hashed assets, no-cache for index.html. MAIN_FILE = "/app/backend/main.py" with open(MAIN_FILE) as f: main_content = f.read() static_import_old = "from fastapi.staticfiles import StaticFiles" static_import_new = """from fastapi.staticfiles import StaticFiles class _CachedStaticFiles(StaticFiles): \"\"\"StaticFiles with safe cache headers. - Hashed build assets (vite emits index-.js) get a 1-year immutable Cache-Control so repeat visits skip the download. - index.html and everything else get no-cache so the browser always revalidates and never pins a stale HTML that references a bundle hash that no longer exists after a rebuild (blank-page 404 bug). \"\"\" async def get_response(self, path: str, scope): resp = await super().get_response(path, scope) if scope["method"] == "GET" and resp.status_code == 200: if path.startswith("assets/") or path.endswith( (".js", ".css", ".woff2", ".png", ".webp", ".svg", ".ico", ".jpg", ".jpeg", ".ttf") ): resp.headers["Cache-Control"] = "public, max-age=31536000, immutable" else: resp.headers["Cache-Control"] = "no-cache" return resp""" if static_import_old in main_content: main_content = main_content.replace(static_import_old, static_import_new) else: print("WARN: StaticFiles import not found for cache patch, may already be applied") static_mount_old = 'app.mount("/", StaticFiles(directory=str(static_path), html=True), name="static")' static_mount_new = 'app.mount("/", _CachedStaticFiles(directory=str(static_path), html=True), name="static")' if static_mount_old in main_content: main_content = main_content.replace(static_mount_old, static_mount_new) else: print("WARN: StaticFiles mount not found for cache patch, may already be applied") try: ast.parse(main_content) print("OK: main.py syntax OK") except SyntaxError as e: print(f"FAIL: Syntax error in main.py: {e}") raise with open(MAIN_FILE, "w") as f: f.write(main_content) print("OK: main.py patched - immutable asset cache + no-cache index.html (fixes stale-hash blank page)")