Files
AgentLab/backend/app/api/knowledge.py
T
ThearchyHelios 964eeaa829 AgentLab:可视化 Agent 编排实验台
在画布上拖拽编排 agent 工作流,并实时看着它执行。

后端 FastAPI + LangGraph:把画布 JSON 编译成 StateGraph,15 种节点
(模型/Agent/多Agent、工具/沙箱、分支/循环/子图、记忆/检索/整形、
人工介入/结构校验)。选 LangGraph 是因为可视化编排真正难的是暂停、
恢复和回放,而 checkpointer + interrupt() 正好把这几件事做在了底层。

前端 React + React Flow:节点元数据驱动属性面板与连接桩,运行时
节点高亮、边流动画、卡片内流式 token 全部由同一份事件流驱动。

关键实现:
- Agent 内工具审批用 @task 缓存模型调用,interrupt 重放时不重复计费
- 沙箱保持只读根文件系统,代码经 exec+base64 写入(put_archive 会被
  只读容器拒绝),断网、非 root、内存/CPU/超时限额
- 检索为向量与 BM25 混合,本地哈希 embedding 零配置可用
- 分支条件走 AST 白名单求值器;HTTP 工具逐 IP 拦截内网与元数据地址
- Mock provider 让无 API Key 也能跑通完整链路

7 个内置模板全部冒烟通过,含真实模型端到端验证。
2026-09-13 17:27:00 +08:00

249 lines
7.7 KiB
Python

from __future__ import annotations
from typing import Any
from fastapi import APIRouter, Depends, File, HTTPException, Query, UploadFile
from pydantic import BaseModel, Field
from sqlalchemy import select
from sqlalchemy.ext.asyncio import AsyncSession
from app.db.base import get_session
from app.db.models import Document, MemoryItem, Skill
from app.memory import kb, store
# --------------------------------------------------------------------------
# 长期记忆
# --------------------------------------------------------------------------
memory_router = APIRouter(prefix="/api/memory", tags=["memory"])
class MemoryIn(BaseModel):
content: str = Field(min_length=1)
scope: str = "default"
kind: str = "fact"
importance: float = Field(default=0.5, ge=0, le=1)
meta: dict[str, Any] = Field(default_factory=dict)
class MemoryOut(BaseModel):
id: str
scope: str
kind: str
content: str
importance: float
use_count: int
meta: dict[str, Any]
created_at: Any = None
model_config = {"from_attributes": True}
@memory_router.get("/scopes")
async def memory_scopes(session: AsyncSession = Depends(get_session)) -> list[dict[str, Any]]:
return await store.list_scopes(session)
@memory_router.get("", response_model=list[MemoryOut])
async def list_memory(
scope: str | None = None,
limit: int = Query(default=200, le=500),
session: AsyncSession = Depends(get_session),
) -> list[MemoryItem]:
return await store.list_memories(session, scope=scope, limit=limit)
@memory_router.post("", response_model=MemoryOut, status_code=201)
async def add_memory(
payload: MemoryIn, session: AsyncSession = Depends(get_session)
) -> MemoryItem:
return await store.remember(
session,
scope=payload.scope,
content=payload.content,
kind=payload.kind,
importance=payload.importance,
meta=payload.meta,
)
@memory_router.get("/search")
async def search_memory(
q: str,
scope: str = "default",
limit: int = 10,
session: AsyncSession = Depends(get_session),
) -> dict[str, Any]:
hits = await store.recall(session, scope=scope, query=q, limit=limit)
return {"query": q, "scope": scope, "results": hits}
@memory_router.delete("/{memory_id}", status_code=204)
async def delete_memory(
memory_id: str, session: AsyncSession = Depends(get_session)
) -> None:
if not await store.forget(session, memory_id):
raise HTTPException(404, "记忆不存在")
@memory_router.delete("/scope/{scope}")
async def clear_scope(scope: str, session: AsyncSession = Depends(get_session)) -> dict[str, int]:
return {"removed": await store.clear_scope(session, scope)}
# --------------------------------------------------------------------------
# 知识库
# --------------------------------------------------------------------------
kb_router = APIRouter(prefix="/api/kb", tags=["knowledge"])
class DocumentOut(BaseModel):
id: str
collection: str
title: str
source: str
mime: str
chunk_count: int
created_at: Any = None
model_config = {"from_attributes": True}
@kb_router.get("/collections")
async def collections(session: AsyncSession = Depends(get_session)) -> list[dict[str, Any]]:
return await kb.list_collections(session)
@kb_router.get("/documents", response_model=list[DocumentOut])
async def list_documents(
collection: str | None = None,
session: AsyncSession = Depends(get_session),
) -> list[Document]:
stmt = select(Document).order_by(Document.created_at.desc()).limit(300)
if collection:
stmt = stmt.where(Document.collection == collection)
return list((await session.execute(stmt)).scalars())
class IngestIn(BaseModel):
collection: str = "default"
title: str = ""
content: str = Field(min_length=1)
source: str = ""
@kb_router.post("/documents", response_model=DocumentOut, status_code=201)
async def ingest(payload: IngestIn, session: AsyncSession = Depends(get_session)) -> Document:
return await kb.ingest_document(
session,
collection=payload.collection,
title=payload.title,
content=payload.content,
source=payload.source,
)
@kb_router.post("/upload", response_model=DocumentOut, status_code=201)
async def upload(
file: UploadFile = File(...),
collection: str = "default",
session: AsyncSession = Depends(get_session),
) -> Document:
raw = await file.read()
if len(raw) > 10 * 1024 * 1024:
raise HTTPException(413, "文件超过 10MB")
try:
content = raw.decode("utf-8")
except UnicodeDecodeError:
raise HTTPException(415, "目前只支持 UTF-8 文本文件(txt / md / json / csv 等)") from None
return await kb.ingest_document(
session,
collection=collection,
title=file.filename or "上传文档",
content=content,
source=file.filename or "",
mime=file.content_type or "text/plain",
)
@kb_router.get("/search")
async def search_kb(
q: str,
collection: str | None = None,
limit: int = 5,
alpha: float = Query(default=0.5, ge=0, le=1, description="1=纯向量, 0=纯关键词"),
session: AsyncSession = Depends(get_session),
) -> dict[str, Any]:
hits = await kb.search(session, collection=collection, query=q, limit=limit, alpha=alpha)
return {"query": q, "collection": collection, "results": hits}
@kb_router.delete("/documents/{doc_id}", status_code=204)
async def delete_document(doc_id: str, session: AsyncSession = Depends(get_session)) -> None:
if not await kb.delete_document(session, doc_id):
raise HTTPException(404, "文档不存在")
# --------------------------------------------------------------------------
# Skill(方法论)
# --------------------------------------------------------------------------
skills_router = APIRouter(prefix="/api/skills", tags=["knowledge"])
class SkillIn(BaseModel):
name: str = Field(min_length=1, max_length=100)
description: str = ""
instructions: str = ""
examples: list[dict[str, Any]] = Field(default_factory=list)
suggested_tools: list[str] = Field(default_factory=list)
tags: list[str] = Field(default_factory=list)
enabled: bool = True
class SkillOut(SkillIn):
id: str
created_at: Any = None
model_config = {"from_attributes": True}
@skills_router.get("", response_model=list[SkillOut])
async def list_skills(session: AsyncSession = Depends(get_session)) -> list[Skill]:
return list((await session.execute(select(Skill).order_by(Skill.created_at.desc()))).scalars())
@skills_router.post("", response_model=SkillOut, status_code=201)
async def create_skill(payload: SkillIn, session: AsyncSession = Depends(get_session)) -> Skill:
if (await session.execute(select(Skill).where(Skill.name == payload.name))).scalar_one_or_none():
raise HTTPException(409, "同名 Skill 已存在")
row = Skill(**payload.model_dump())
session.add(row)
await session.commit()
await session.refresh(row)
return row
@skills_router.patch("/{skill_id}", response_model=SkillOut)
async def update_skill(
skill_id: str, payload: SkillIn, session: AsyncSession = Depends(get_session)
) -> Skill:
row = await session.get(Skill, skill_id)
if not row:
raise HTTPException(404, "Skill 不存在")
for key, value in payload.model_dump().items():
setattr(row, key, value)
await session.commit()
await session.refresh(row)
return row
@skills_router.delete("/{skill_id}", status_code=204)
async def delete_skill(skill_id: str, session: AsyncSession = Depends(get_session)) -> None:
row = await session.get(Skill, skill_id)
if not row:
raise HTTPException(404, "Skill 不存在")
await session.delete(row)
await session.commit()