From b9f5e72bbea72e85d4c5e10b1d469414b17cd754 Mon Sep 17 00:00:00 2001 From: banxian1987 Date: Thu, 27 Aug 2026 00:00:53 +0800 Subject: [PATCH] feat(server): auto-load .env and resolve channels with robust fallback --- strix_hub/server.py | 149 ++++++++++++++++++++++---------------- strix_hub/task_manager.py | 34 +++++++-- 2 files changed, 115 insertions(+), 68 deletions(-) diff --git a/strix_hub/server.py b/strix_hub/server.py index e0cab209..37c44ca8 100644 --- a/strix_hub/server.py +++ b/strix_hub/server.py @@ -23,68 +23,95 @@ logger = logging.getLogger("strix_hub.server") STATIC_DIR = Path(__file__).resolve().parent / "static" -# Configurable Local LLM environment (reads from env vars or defaults to generic placeholders) -LOCAL_LLM_MODEL = os.environ.get("LOCAL_LLM_MODEL", "openai/Qwen3.8-27B-abliterated") -LOCAL_LLM_URL = os.environ.get("LOCAL_LLM_URL", os.environ.get("OPENAI_BASE_URL", "http://127.0.0.1:8000/v1")) -LOCAL_LLM_KEY = os.environ.get("LOCAL_LLM_KEY", os.environ.get("OPENAI_API_KEY", "")) +# Automatically load .env file from working directory or /opt/strix/.env if available +def _load_env_file() -> None: + for env_path in [Path.cwd() / ".env", Path("/opt/strix/.env")]: + if env_path.is_file(): + try: + with open(env_path, "r", encoding="utf-8") as f: + for line in f: + line = line.strip() + if not line or line.startswith("#") or "=" not in line: + continue + k, v = line.split("=", 1) + k, v = k.strip(), v.strip().strip("'\"") + if k and k not in os.environ: + os.environ[k] = v + except Exception: + pass -MODEL_PRESETS = [ - { - "id": "hybrid-gemini-qwen", - "name": "🚀 顶配混合动力 (Gemini 3.1 Pro 大脑 + 本地 Qwen 3.8 打手)", - "root_model": "openai/gemini-3.1-pro-preview", - "root_api_base": "", - "root_api_key": "", - "subagent_model": LOCAL_LLM_MODEL, - "subagent_api_base": LOCAL_LLM_URL, - "subagent_api_key": LOCAL_LLM_KEY, - "description": "【最佳推荐】主控用云端 Gemini 3.1 Pro 百万上下文做复杂漏洞挖掘；海量并发子智能体全部走本地私有化模型，零成本无外网限流！", - }, - { - "id": "local-pure-cluster", - "name": "🛡️ 本地全离线集群 (主子全跑本地私有化模型)", - "root_model": LOCAL_LLM_MODEL, - "root_api_base": LOCAL_LLM_URL, - "root_api_key": LOCAL_LLM_KEY, - "subagent_model": LOCAL_LLM_MODEL, - "subagent_api_base": LOCAL_LLM_URL, - "subagent_api_key": LOCAL_LLM_KEY, - "description": "完全在企业局域网内运行，数据绝不出网，适合离线环境与内网安全合规审计。", - }, - { - "id": "gemini-optimal", - "name": "⚡ Gemini 纯云端组合 (3.1 Pro + 3.5 Flash)", - "root_model": "openai/gemini-3.1-pro-preview", - "root_api_base": "", - "root_api_key": "", - "subagent_model": "openai/gemini-3.5-flash", - "subagent_api_base": "", - "subagent_api_key": "", - "description": "主控用 3.1 Pro 推理，子任务用 3.5 Flash 极速响应，全云端中转组合。", - }, - { - "id": "claude-hybrid", - "name": "💎 Claude 3.7 安全审计 + 本地模型混合调度", - "root_model": "openai/claude-3-7-sonnet", - "root_api_base": "", - "root_api_key": "", - "subagent_model": LOCAL_LLM_MODEL, - "subagent_api_base": LOCAL_LLM_URL, - "subagent_api_key": LOCAL_LLM_KEY, - "description": "主控使用顶级安全审计模型 Claude 3.7，子任务由本地私有化集群并发执行。", - }, - { - "id": "custom", - "name": "⚙️ 自定义独立双渠道 (Custom Dual Channels)", - "root_model": "", - "root_api_base": "", - "root_api_key": "", - "subagent_model": "", - "subagent_api_base": "", - "subagent_api_key": "", - "description": "自由为两个模型分别配置不同的 Base URL 与 API Key 渠道。", - }, -] +_load_env_file() + +def get_local_llm_config() -> tuple[str, str, str]: + """Resolve local LLM model name, API URL, and key with rich fallbacks.""" + model = os.environ.get("LOCAL_LLM_MODEL", os.environ.get("STRIX_LLM", "openai/Qwen3.8-27B-abliterated")) + url = os.environ.get("LOCAL_LLM_URL", os.environ.get("LLM_API_BASE", os.environ.get("OPENAI_BASE_URL", "http://127.0.0.1:8000/v1"))) + key = os.environ.get("LOCAL_LLM_KEY", os.environ.get("LLM_API_KEY", os.environ.get("OPENAI_API_KEY", ""))) + return model, url, key + +LOCAL_LLM_MODEL, LOCAL_LLM_URL, LOCAL_LLM_KEY = get_local_llm_config() + +def get_model_presets() -> list[dict[str, Any]]: + m, u, k = get_local_llm_config() + return [ + { + "id": "hybrid-gemini-qwen", + "name": "🚀 顶配混合动力 (Gemini 3.1 Pro 大脑 + 本地 Qwen 3.8 打手)", + "root_model": "openai/gemini-3.1-pro-preview", + "root_api_base": "", + "root_api_key": "", + "subagent_model": m, + "subagent_api_base": u, + "subagent_api_key": k, + "description": "【最佳推荐】主控用云端 Gemini 3.1 Pro 百万上下文做复杂漏洞挖掘；海量并发子智能体全部走本地私有化模型，零成本无外网限流！", + }, + { + "id": "local-pure-cluster", + "name": "🛡️ 本地全离线集群 (主子全跑本地私有化模型)", + "root_model": m, + "root_api_base": u, + "root_api_key": k, + "subagent_model": m, + "subagent_api_base": u, + "subagent_api_key": k, + "description": "完全在企业局域网内运行，数据绝不出网，适合离线环境与内网安全合规审计。", + }, + { + "id": "gemini-optimal", + "name": "⚡ Gemini 纯云端组合 (3.1 Pro + 3.5 Flash)", + "root_model": "openai/gemini-3.1-pro-preview", + "root_api_base": "", + "root_api_key": "", + "subagent_model": "openai/gemini-3.5-flash", + "subagent_api_base": "", + "subagent_api_key": "", + "description": "主控用 3.1 Pro 推理，子任务用 3.5 Flash 极速响应，全云端中转组合。", + }, + { + "id": "claude-hybrid", + "name": "💎 Claude 3.7 安全审计 + 本地模型混合调度", + "root_model": "openai/claude-3-7-sonnet", + "root_api_base": "", + "root_api_key": "", + "subagent_model": m, + "subagent_api_base": u, + "subagent_api_key": k, + "description": "主控使用顶级安全审计模型 Claude 3.7，子任务由本地私有化集群并发执行。", + }, + { + "id": "custom", + "name": "⚙️ 自定义独立双渠道 (Custom Dual Channels)", + "root_model": "", + "root_api_base": "", + "root_api_key": "", + "subagent_model": "", + "subagent_api_base": "", + "subagent_api_key": "", + "description": "自由为两个模型分别配置不同的 Base URL 与 API Key 渠道。", + }, + ] + +MODEL_PRESETS = get_model_presets() def make_hub_handler() -> type[BaseHTTPRequestHandler]: diff --git a/strix_hub/task_manager.py b/strix_hub/task_manager.py index 04b67870..b98a3ffe 100644 --- a/strix_hub/task_manager.py +++ b/strix_hub/task_manager.py @@ -60,15 +60,35 @@ def start_task(task_id: str) -> dict[str, Any]: # 1. Start dedicated Dual-Channel ModelRouter on a free port for this task port = 18800 + (abs(hash(task_id)) % 1000) - # Fallback to server env if task channel is left blank - default_base = os.environ.get("OPENAI_BASE_URL") or os.environ.get("LLM_API_BASE", "https://api.openai.com/v1") - default_key = os.environ.get("OPENAI_API_KEY") or os.environ.get("LLM_API_KEY", "") + # Fallback to server env if task channel is left blank or has unreachable placeholder 8000 + env_base = ( + os.environ.get("LOCAL_LLM_URL") + or os.environ.get("OPENAI_BASE_URL") + or os.environ.get("LLM_API_BASE", "") + ) + env_key = ( + os.environ.get("LOCAL_LLM_KEY") + or os.environ.get("OPENAI_API_KEY") + or os.environ.get("LLM_API_KEY", "") + ) - root_base = task.get("root_api_base") or task.get("api_base") or default_base - root_key = task.get("root_api_key_raw") or task.get("api_key_raw") or default_key + def _resolve_channel(base: str | None, key: str | None) -> tuple[str, str]: + b = (base or "").strip() + k = (key or "").strip() + if (not b or "127.0.0.1:8000" in b) and env_base: + b = env_base + if not k and env_key: + k = env_key + return b, k - sub_base = task.get("subagent_api_base") or task.get("api_base") or default_base - sub_key = task.get("subagent_api_key_raw") or task.get("api_key_raw") or default_key + root_base, root_key = _resolve_channel( + task.get("root_api_base") or task.get("api_base"), + task.get("root_api_key_raw") or task.get("api_key_raw"), + ) + sub_base, sub_key = _resolve_channel( + task.get("subagent_api_base") or task.get("api_base"), + task.get("subagent_api_key_raw") or task.get("api_key_raw"), + ) router = ModelRouterServer( host="127.0.0.1",