Files
planet/aiprovider/provider_service.py
rayd1o cee1996809
Some checks failed
ci / backend (push) Has been cancelled
ci / frontend (push) Has been cancelled
release / images (push) Has been cancelled
ci / delivery (push) Has been cancelled
release: bump version to 0.74.5
2026-09-13 14:04:34 +08:00

489 lines
19 KiB
Python

from __future__ import annotations
import asyncio
import json
from typing import Any
from uuid import NAMESPACE_URL, uuid4, uuid5
import httpx
from fastapi import HTTPException, status
from aiprovider.config import settings
from aiprovider.schemas import (
AIContentBlock,
AIProviderStatusResponse,
SituationalAnalysisRequest,
SituationalAnalysisResponse,
)
def _normalize_provider(value: str) -> str:
return (value or "disabled").strip().lower()
def _normalize_provider_api(value: str) -> str:
return (value or "auto").strip().lower().replace("_", "-")
def _resolve_provider_api(provider: str, configured_api: str) -> str:
if configured_api and configured_api != "auto":
return configured_api
if provider in {"openai", "openai-compatible", "openai_compatible"}:
return "openai-completions"
if provider in {
"anthropic",
"anthropic-compatible",
"anthropic_compatible",
"claude-compatible",
"claude_compatible",
"minimax",
"kimi-coding",
"moonshot-anthropic",
}:
return "anthropic-messages"
if provider == "ollama":
return "ollama-generate"
return "disabled"
class ProviderService:
def __init__(self, overrides: dict[str, Any] | None = None) -> None:
overrides = overrides or {}
self.provider = _normalize_provider(overrides.get("provider") or settings.AI_PROVIDER)
self.provider_api = _resolve_provider_api(
self.provider,
_normalize_provider_api(overrides.get("provider_api") or settings.AI_PROVIDER_API),
)
self.base_url = str(overrides.get("base_url") or settings.AI_BASE_URL).rstrip("/")
self.api_key = str(overrides.get("api_key") or settings.AI_API_KEY)
self.default_model = str(overrides.get("model") or settings.AI_MODEL)
self.timeout = settings.AI_TIMEOUT_SECONDS
self.http_retry_attempts = max(settings.AI_HTTP_RETRY_ATTEMPTS, 1)
self.max_tokens = int(overrides.get("max_tokens") or settings.AI_MAX_TOKENS)
self.anthropic_version = str(
overrides.get("anthropic_version") or settings.AI_ANTHROPIC_VERSION
)
self.model_provider_apis = self._parse_model_provider_apis(
overrides.get("model_provider_apis")
)
self.session_id = str(uuid4())
def get_status(self) -> AIProviderStatusResponse:
enabled = self.provider != "disabled"
has_credentials = bool(self.api_key) if self._requires_api_key() else True
configured = enabled and bool(self.base_url and has_credentials and self.default_model)
return AIProviderStatusResponse(
provider=self.provider,
api=self.provider_api if enabled else None,
enabled=enabled,
configured=configured,
model=self.default_model or None,
base_url=self.base_url if enabled else None,
)
async def analyze(self, payload: SituationalAnalysisRequest) -> SituationalAnalysisResponse:
if self.provider == "disabled":
raise HTTPException(
status_code=status.HTTP_503_SERVICE_UNAVAILABLE,
detail="AI provider is disabled. Configure AI_PROVIDER in .env to enable analysis.",
)
model = payload.preferred_model or self.default_model
has_credentials = bool(self.api_key) if self._requires_api_key() else True
if not self.base_url or not has_credentials or not model:
raise HTTPException(
status_code=status.HTTP_503_SERVICE_UNAVAILABLE,
detail="AI provider is not fully configured. Check AI_BASE_URL, AI_API_KEY, and AI_MODEL.",
)
prompt = self._build_prompt(payload)
if payload.context.get("session_id") is not None:
self.session_id = str(uuid5(NAMESPACE_URL, f"planet:{payload.context['session_id']}"))
provider_api = self._resolve_model_provider_api(model)
if provider_api == "openai-completions":
data = await self._request_openai_compatible(model, prompt, payload.system_prompt)
content = self._extract_openai_content(data)
content_blocks = self._extract_openai_blocks(data)
elif provider_api == "openai-responses":
data = await self._request_openai_responses(model, prompt, payload.system_prompt)
content_blocks = self._extract_responses_blocks(data)
content = "".join(block.text for block in content_blocks if block.text)
elif provider_api == "anthropic-messages":
data = await self._request_anthropic_messages(
model,
prompt,
payload.thinking,
payload.system_prompt,
)
content = self._extract_anthropic_content(data)
content_blocks = self._extract_anthropic_blocks(data)
elif provider_api == "ollama-generate":
data = await self._request_ollama(model, prompt, payload.system_prompt)
content = self._extract_ollama_content(data)
content_blocks = self._extract_ollama_blocks(data)
else:
raise HTTPException(
status_code=status.HTTP_400_BAD_REQUEST,
detail=f"Unsupported AI provider API: {self.provider_api}",
)
text_blocks = [block.text for block in content_blocks if block.text]
thinking_blocks = [block.thinking for block in content_blocks if block.thinking]
return SituationalAnalysisResponse(
provider=self.provider,
model=model,
content=content,
content_blocks=content_blocks,
text_blocks=text_blocks,
thinking_blocks=thinking_blocks,
raw_response=data,
)
def _requires_api_key(self) -> bool:
return self.provider_api != "ollama-generate"
def _resolve_model_provider_api(self, model: str) -> str:
return self.model_provider_apis.get(model) or self.provider_api
def _parse_model_provider_apis(self, value: Any) -> dict[str, str]:
if isinstance(value, dict):
raw = value
elif isinstance(value, str) and value.strip():
try:
parsed = json.loads(value)
except json.JSONDecodeError:
return {}
raw = parsed if isinstance(parsed, dict) else {}
else:
raw = {}
return {
str(model): _normalize_provider_api(str(provider_api))
for model, provider_api in raw.items()
if model and provider_api
}
def _build_prompt(self, payload: SituationalAnalysisRequest) -> str:
sections = [
f"任务标题:\n{payload.title}",
f"分析目标:\n{payload.objective}",
]
if payload.observations:
sections.append("观测事实:\n" + "\n".join(f"- {item}" for item in payload.observations))
if payload.constraints:
sections.append("约束条件:\n" + "\n".join(f"- {item}" for item in payload.constraints))
if payload.context:
sections.append(f"附加上下文:\n{payload.context}")
return "\n\n".join(sections)
def _resolve_system_prompt(self, system_prompt: str | None) -> str | None:
resolved = str(system_prompt or "").strip()
return resolved or None
async def _request_openai_compatible(
self,
model: str,
prompt: str,
system_prompt: str | None = None,
) -> dict[str, Any]:
messages = []
resolved_system_prompt = self._resolve_system_prompt(system_prompt)
if resolved_system_prompt:
messages.append({"role": "system", "content": resolved_system_prompt})
messages.append({"role": "user", "content": prompt})
request_body = {
"model": model,
"messages": messages,
"temperature": 0.2,
"max_tokens": self.max_tokens,
}
return await self._post(
path="/chat/completions",
headers={
"Authorization": f"Bearer {self.api_key}",
"Content-Type": "application/json",
},
request_body=request_body,
)
async def _request_openai_responses(
self, model: str, prompt: str, system_prompt: str | None = None,
) -> dict[str, Any]:
request_body: dict[str, Any] = {
"model": model, "input": prompt, "max_output_tokens": self.max_tokens, "store": False,
}
resolved_system_prompt = self._resolve_system_prompt(system_prompt)
if resolved_system_prompt:
request_body["instructions"] = resolved_system_prompt
return await self._post(
path="/responses",
headers={"Authorization": f"Bearer {self.api_key}", "Content-Type": "application/json"},
request_body=request_body,
)
def _extract_responses_blocks(self, payload: dict[str, Any]) -> list[AIContentBlock]:
blocks: list[AIContentBlock] = []
for item in payload.get("output") or []:
if not isinstance(item, dict):
continue
if item.get("type") == "message":
for part in item.get("content") or []:
if not isinstance(part, dict):
continue
text = part.get("text") or part.get("refusal")
if isinstance(text, str) and text:
blocks.append(AIContentBlock(type="text", text=text))
elif item.get("type") == "reasoning":
for part in item.get("summary") or []:
if isinstance(part, dict) and isinstance(part.get("text"), str):
blocks.append(AIContentBlock(type="thinking", thinking=part["text"]))
return blocks
async def _request_anthropic_messages(
self,
model: str,
prompt: str,
thinking: dict[str, Any] | None = None,
system_prompt: str | None = None,
) -> dict[str, Any]:
request_body = {
"model": model,
"messages": [
{
"role": "user",
"content": [
{
"type": "text",
"text": prompt,
}
],
}
],
"max_tokens": self.max_tokens,
"temperature": 0.2,
}
resolved_system_prompt = self._resolve_system_prompt(system_prompt)
if resolved_system_prompt:
request_body["system"] = resolved_system_prompt
resolved_thinking = self._resolve_anthropic_thinking(thinking, model)
if resolved_thinking:
request_body["thinking"] = resolved_thinking
if self.provider == "minimax" and self.base_url.endswith("/anthropic"):
path = "/v1/messages"
else:
path = "/messages"
return await self._post(
path=path,
headers={
"x-api-key": self.api_key,
"anthropic-version": self.anthropic_version,
"Content-Type": "application/json",
},
request_body=request_body,
)
def _resolve_anthropic_thinking(
self, thinking: dict[str, Any] | None, model: str,
) -> dict[str, Any] | None:
if thinking:
if model.casefold() == "minimax-m3" and thinking.get("type") == "enabled":
return {"type": "adaptive"}
return thinking
# OpenClaw treats MiniMax's Anthropic-compatible path specially:
# disable thinking by default unless the caller explicitly opts in.
if self.provider == "minimax":
return {"type": "disabled"}
return None
async def _request_anthropic_compatible(
self,
model: str,
prompt: str,
thinking: dict[str, Any] | None = None,
system_prompt: str | None = None,
) -> dict[str, Any]:
return await self._request_anthropic_messages(model, prompt, thinking, system_prompt)
async def _request_ollama(
self,
model: str,
prompt: str,
system_prompt: str | None = None,
) -> dict[str, Any]:
request_body = {
"model": model,
"stream": False,
"prompt": prompt,
"options": {
"temperature": 0.2,
"num_predict": self.max_tokens,
},
}
resolved_system_prompt = self._resolve_system_prompt(system_prompt)
if resolved_system_prompt:
request_body["system"] = resolved_system_prompt
return await self._post(
path="/api/generate",
headers={
"Content-Type": "application/json",
},
request_body=request_body,
)
async def _post(
self,
path: str,
headers: dict[str, str],
request_body: dict[str, Any],
) -> dict[str, Any]:
headers = {"User-Agent": "Planet/1.0", **headers}
if self.provider == "opencode-go":
headers["x-opencode-session"] = self.session_id
last_error: Exception | None = None
for attempt in range(1, self.http_retry_attempts + 1):
try:
async with httpx.AsyncClient(timeout=self.timeout) as client:
response = await client.post(
f"{self.base_url}{path}",
headers=headers,
json=request_body,
)
response.raise_for_status()
return response.json()
except httpx.HTTPStatusError as exc:
last_error = exc
if attempt < self.http_retry_attempts and exc.response.status_code >= 500:
await asyncio.sleep(0.3 * attempt)
continue
detail = exc.response.text or "AI provider returned an error"
raise HTTPException(
status_code=status.HTTP_502_BAD_GATEWAY,
detail=f"AI provider request failed: {detail}",
) from exc
except httpx.HTTPError as exc:
last_error = exc
if attempt < self.http_retry_attempts:
await asyncio.sleep(0.3 * attempt)
continue
raise HTTPException(
status_code=status.HTTP_502_BAD_GATEWAY,
detail=f"Failed to reach AI provider: {exc}",
) from exc
raise HTTPException(
status_code=status.HTTP_502_BAD_GATEWAY,
detail=f"AI provider request failed: {last_error}",
)
def _extract_openai_content(self, payload: dict[str, Any]) -> str:
choices = payload.get("choices") or []
if not choices:
return ""
message = choices[0].get("message") or {}
content = message.get("content")
if isinstance(content, str):
if content:
return content
reasoning_content = message.get("reasoning_content")
return reasoning_content if isinstance(reasoning_content, str) else ""
if isinstance(content, list):
return "".join(
item.get("text", "")
for item in content
if isinstance(item, dict)
)
reasoning_content = message.get("reasoning_content")
if isinstance(reasoning_content, str):
return reasoning_content
return ""
def _extract_openai_blocks(self, payload: dict[str, Any]) -> list[AIContentBlock]:
choices = payload.get("choices") or []
if not choices:
return []
message = choices[0].get("message") or {}
content = message.get("content")
if isinstance(content, str):
blocks = [AIContentBlock(type="text", text=content)] if content else []
reasoning_content = message.get("reasoning_content")
if isinstance(reasoning_content, str) and reasoning_content:
blocks.append(AIContentBlock(type="thinking", thinking=reasoning_content))
return blocks
if not isinstance(content, list):
reasoning_content = message.get("reasoning_content")
return [AIContentBlock(type="thinking", thinking=reasoning_content)] if isinstance(reasoning_content, str) and reasoning_content else []
blocks: list[AIContentBlock] = []
for item in content:
if not isinstance(item, dict):
continue
blocks.append(
AIContentBlock(
type=str(item.get("type", "text")),
text=item.get("text") if isinstance(item.get("text"), str) else None,
metadata={k: v for k, v in item.items() if k not in {"type", "text"}},
)
)
reasoning_content = message.get("reasoning_content")
if isinstance(reasoning_content, str) and reasoning_content:
blocks.append(AIContentBlock(type="thinking", thinking=reasoning_content))
return blocks
def _extract_anthropic_content(self, payload: dict[str, Any]) -> str:
content = payload.get("content")
if isinstance(content, str):
return content
if not isinstance(content, list):
return ""
fragments: list[str] = []
for item in content:
if not isinstance(item, dict):
continue
if item.get("type") == "text" and isinstance(item.get("text"), str):
fragments.append(item["text"])
return "".join(fragments)
def _extract_anthropic_blocks(self, payload: dict[str, Any]) -> list[AIContentBlock]:
content = payload.get("content")
if not isinstance(content, list):
return []
blocks: list[AIContentBlock] = []
for item in content:
if not isinstance(item, dict):
continue
blocks.append(
AIContentBlock(
type=str(item.get("type", "unknown")),
text=item.get("text") if isinstance(item.get("text"), str) else None,
thinking=item.get("thinking") if isinstance(item.get("thinking"), str) else None,
signature=item.get("signature") if isinstance(item.get("signature"), str) else None,
metadata={
k: v
for k, v in item.items()
if k not in {"type", "text", "thinking", "signature"}
},
)
)
return blocks
def _extract_ollama_content(self, payload: dict[str, Any]) -> str:
response = payload.get("response")
if isinstance(response, str):
return response
return ""
def _extract_ollama_blocks(self, payload: dict[str, Any]) -> list[AIContentBlock]:
response = payload.get("response")
if isinstance(response, str) and response:
return [AIContentBlock(type="text", text=response)]
return []