0e5368a894
- llm_bridge 新增 llm_call_msgs_native 支持 tools 参数+解析 tool_calls - agent_loop_v2._call_llm 优先走 native function calling，失败回退文本 - run loop 处理原生 tool_calls（role=tool 回填） - _init_components 把 config.tools 注册进 ToolRegistry（修复 registry 空导致 tools_description 为空、schema 为空的 bug）
207 lines
7.7 KiB
Python
207 lines
7.7 KiB
Python
"""
|
||
LLM bridge for pipeline handlers.
|
||
|
||
Provides a simple async interface for handlers to call LLM APIs.
|
||
Looks up model config from the llm database table first,
|
||
falls back to environment variables.
|
||
"""
|
||
|
||
import json
|
||
import logging
|
||
import os
|
||
|
||
logger = logging.getLogger("pipeline.llm_bridge")
|
||
|
||
# Cache: model_name -> {api_base, api_key, model_id}
|
||
_model_cache: dict = {}
|
||
|
||
|
||
def _decrypt_key(encrypted: str) -> str:
|
||
"""Decrypt api_key stored with password_encode. Falls back to plaintext."""
|
||
if not encrypted:
|
||
return ""
|
||
try:
|
||
from appPublic.rc4 import unpassword
|
||
from appPublic.jsonConfig import getConfig
|
||
config = getConfig()
|
||
key = config.password_key
|
||
return unpassword(key, encrypted)
|
||
except Exception:
|
||
return encrypted # already plaintext or decrypt failed
|
||
|
||
|
||
async def _get_model_config(model_name: str = None) -> dict:
|
||
"""Look up model config from llm table. Returns dict with api_base, api_key, model_id."""
|
||
global _model_cache
|
||
|
||
if model_name and model_name in _model_cache:
|
||
return _model_cache[model_name]
|
||
|
||
try:
|
||
from sqlor.dbpools import DBPools
|
||
db = DBPools()
|
||
dbname = "pipeline"
|
||
async with db.sqlorContext(dbname) as sor:
|
||
if model_name:
|
||
sql = "SELECT api_base, api_key, model_id FROM llm WHERE name=${name}$ AND status='active' LIMIT 1"
|
||
recs = await sor.sqlExe(sql, {"name": model_name})
|
||
else:
|
||
sql = "SELECT api_base, api_key, model_id, name FROM llm WHERE status='active' ORDER BY id LIMIT 1"
|
||
recs = await sor.sqlExe(sql, {})
|
||
if recs:
|
||
r = recs[0]
|
||
cfg = {
|
||
"api_base": getattr(r, "api_base", "") or "",
|
||
"api_key": _decrypt_key(getattr(r, "api_key", "") or ""),
|
||
"model_id": getattr(r, "model_id", "") or "",
|
||
}
|
||
cache_key = model_name or getattr(r, "name", "")
|
||
if cache_key:
|
||
_model_cache[cache_key] = cfg
|
||
return cfg
|
||
except Exception as e:
|
||
logger.warning("llm_bridge: DB lookup failed: %s", e)
|
||
|
||
return {}
|
||
|
||
|
||
async def llm_call(prompt: str, model: str = None, temperature: float = 0.7) -> str:
|
||
"""Call LLM and return text response.
|
||
|
||
Backend priority:
|
||
1. harnessed_agent.llm_chat (if loaded in ServerEnv)
|
||
2. DB llm table (api_base + api_key)
|
||
3. Environment variables (LLM_API_BASE, LLM_API_KEY, LLM_MODEL)
|
||
"""
|
||
# Priority 1: harnessed_agent
|
||
try:
|
||
from ahserver.serverenv import ServerEnv
|
||
env = ServerEnv()
|
||
if hasattr(env, 'llm_chat'):
|
||
result = await env.llm_chat(prompt, model=model, temperature=temperature)
|
||
if isinstance(result, dict):
|
||
return result.get("content", result.get("text", str(result)))
|
||
return str(result)
|
||
except Exception:
|
||
pass
|
||
|
||
# Priority 2: DB llm table
|
||
cfg = await _get_model_config(model)
|
||
if cfg.get("api_key") and cfg.get("api_base"):
|
||
api_base = cfg["api_base"]
|
||
api_key = cfg["api_key"]
|
||
model_id = cfg.get("model_id") or model or "default"
|
||
logger.info("llm_bridge: using DB model config for %s -> %s", model, api_base)
|
||
else:
|
||
# Priority 3: Environment variables
|
||
api_base = os.environ.get("LLM_API_BASE", "https://api.openai.com/v1")
|
||
api_key = os.environ.get("LLM_API_KEY", "")
|
||
model_id = model or os.environ.get("LLM_MODEL", "gpt-4o-mini")
|
||
|
||
if not api_key:
|
||
raise ValueError("No LLM API configured. Please add a model in the llm table or set LLM_API_KEY env var.")
|
||
|
||
import aiohttp
|
||
|
||
headers = {
|
||
"Authorization": f"Bearer {api_key}",
|
||
"Content-Type": "application/json",
|
||
}
|
||
payload = {
|
||
"model": model_id,
|
||
"messages": [{"role": "user", "content": prompt}],
|
||
"temperature": temperature,
|
||
}
|
||
|
||
url = api_base.rstrip("/") + "/chat/completions"
|
||
async with aiohttp.ClientSession() as session:
|
||
async with session.post(
|
||
url, headers=headers, json=payload,
|
||
timeout=aiohttp.ClientTimeout(total=120)
|
||
) as resp:
|
||
if resp.status != 200:
|
||
text = await resp.text()
|
||
raise ValueError(f"LLM API error {resp.status}: {text[:300]}")
|
||
data = await resp.json()
|
||
return data["choices"][0]["message"]["content"]
|
||
|
||
|
||
async def call_llm(tenant_id: str, prompt: str, model: str = None, temperature: float = 0.7) -> str:
|
||
"""SDLC handler interface — delegates to llm_call."""
|
||
return await llm_call(prompt, model=model, temperature=temperature)
|
||
|
||
|
||
async def llm_call_msgs(messages: list, model: str = None, temperature: float = 0.7) -> str:
|
||
"""Call LLM with full message array (system/user/assistant)."""
|
||
import aiohttp
|
||
|
||
cfg = await _get_model_config(model)
|
||
if cfg.get("api_key") and cfg.get("api_base"):
|
||
api_base = cfg["api_base"]
|
||
api_key = cfg["api_key"]
|
||
model_id = cfg.get("model_id") or model or "default"
|
||
else:
|
||
api_base = os.environ.get("LLM_API_BASE", "https://api.openai.com/v1")
|
||
api_key = os.environ.get("LLM_API_KEY", "")
|
||
model_id = model or os.environ.get("LLM_MODEL", "gpt-4o-mini")
|
||
|
||
if not api_key:
|
||
raise ValueError("No LLM API configured")
|
||
|
||
headers = {"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"}
|
||
payload = {"model": model_id, "messages": messages, "temperature": temperature}
|
||
|
||
url = api_base.rstrip("/") + "/chat/completions"
|
||
async with aiohttp.ClientSession() as session:
|
||
async with session.post(url, headers=headers, json=payload,
|
||
timeout=aiohttp.ClientTimeout(total=180)) as resp:
|
||
if resp.status != 200:
|
||
text = await resp.text()
|
||
raise ValueError(f"LLM API error {resp.status}: {text[:300]}")
|
||
data = await resp.json()
|
||
return data["choices"][0]["message"]["content"]
|
||
|
||
|
||
async def llm_call_msgs_native(messages: list, tools: list = None, model: str = None, temperature: float = 0.7) -> dict:
|
||
"""Native function calling. 传入 tools JSON schema，返回 message dict。
|
||
|
||
Returns:
|
||
{"content": str, "tool_calls": [{"id","type","function":{"name","arguments"}}]}
|
||
当模型返回 tool_calls 时，content 通常为空字符串。
|
||
"""
|
||
import aiohttp
|
||
|
||
cfg = await _get_model_config(model)
|
||
if cfg.get("api_key") and cfg.get("api_base"):
|
||
api_base = cfg["api_base"]
|
||
api_key = cfg["api_key"]
|
||
model_id = cfg.get("model_id") or model or "default"
|
||
else:
|
||
api_base = os.environ.get("LLM_API_BASE", "https://api.openai.com/v1")
|
||
api_key = os.environ.get("LLM_API_KEY", "")
|
||
model_id = model or os.environ.get("LLM_MODEL", "gpt-4o-mini")
|
||
|
||
if not api_key:
|
||
raise ValueError("No LLM API configured")
|
||
|
||
headers = {"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"}
|
||
payload = {"model": model_id, "messages": messages, "temperature": temperature}
|
||
if tools:
|
||
payload["tools"] = tools
|
||
payload["tool_choice"] = "auto"
|
||
|
||
url = api_base.rstrip("/") + "/chat/completions"
|
||
async with aiohttp.ClientSession() as session:
|
||
async with session.post(url, headers=headers, json=payload,
|
||
timeout=aiohttp.ClientTimeout(total=180)) as resp:
|
||
if resp.status != 200:
|
||
text = await resp.text()
|
||
raise ValueError(f"LLM API error {resp.status}: {text[:300]}")
|
||
data = await resp.json()
|
||
msg = data["choices"][0]["message"]
|
||
return {
|
||
"content": msg.get("content") or "",
|
||
"tool_calls": msg.get("tool_calls") or [],
|
||
}
|
||
|