from deepagents import create_deep_agent research_agent = { "name": "researcher", "description": "Deep research on any topic using web search", "system_prompt": """You are a research specialist. Use web search to gather comprehensive information. Always cite your sources. Return a structured summary of findings.""", "tools": [web_search], "model": "openai:gpt-4o" } agent = create_deep_agent( subagents=[research_agent] )
CompiledSubAgent 的唯一要求是：返回的 state 里必须有 messages 键，因为结果要从 messages 里提取。
这种类型适合复用已有的 Agent 实现，或者做一些 DeepAgents 默认不支持的高级定制。
AsyncSubAgent：异步子代理
前两种都是同步的：主代理调用子代理，然后等它返回。但有时候子任务执行时间很长，你不想在那干等着。
AsyncSubAgent 支持后台执行、非阻塞调用。
from deepagents import AsyncSubAgent async_subagent = AsyncSubAgent( name="long-running-task", description="Tasks that take a long time", graph_id="my-deployment-id", # LangSmith 部署 ID url="https://my-agent-server.com", headers={"Authorization": "Bearer token"} ) agent = create_deep_agent( subagents=[async_subagent] )
Launch an ephemeral subagent to handle complex, multi-step independent tasks with isolated context windows. When to use the task tool: - When a task is complex and multi-step, and can be fully delegated in isolation - When a task is independent of other tasks and can run in parallel - When a task requires focused reasoning or heavy token/context usage that would bloat the orchestrator thread - When sandboxing improves reliability - When you only care about the output of the subagent, and not the intermediate steps
然后又告诉什么时候不该用：
When NOT to use the task tool: - If you need to see the intermediate reasoning or steps after the subagent has completed - If the task is trivial (a few tool calls or simple lookup) - If delegating does not reduce token usage, complexity, or context switching - If splitting would add latency without benefit
Subagent lifecycle: 1. Spawn → Provide clear role, instructions, and expected output 2. Run → The subagent completes the task autonomously 3. Return → The subagent provides a single structured result 4. Reconcile → Incorporate or synthesize the result into the main thread
from deepagents import create_deep_agent, SubAgent from langchain.tools import tool @tool def web_search(query: str) -> str: """Search the web for information""" # 实现省略 return results # 研究子代理 research_subagent = SubAgent( name="policy-researcher", description="Research AI policies and regulations using web search", system_prompt="""You are a policy research specialist. Research the given topic thoroughly: 1. Search for official documents and sources 2. Find recent news and analysis 3. Compare different perspectives 4. Cite all sources with URLs Return a comprehensive research summary.""", tools=[web_search], model="openai:gpt-4o" ) # 分析子代理 analysis_subagent = SubAgent( name="policy-analyst", description="Analyze policy implications and compare approaches", system_prompt="""You are a policy analyst. Analyze the provided research: 1. Identify key stakeholders and their positions 2. Compare different regulatory approaches 3. Assess potential impacts (economic, social, technical) 4. Highlight areas of consensus and controversy Provide structured analysis.""", tools=[] # 这个子代理不需要工具，只做分析 ) # 写作子代理 writer_subagent = SubAgent( name="report-writer", description="Write professional policy reports", system_prompt="""You are a professional policy report writer. Write a clear, structured report: - Executive summary - Background and context - Key findings - Analysis and implications - Recommendations - References Use formal but accessible language.""", tools=[] ) # 审核子代理 critique_subagent = SubAgent( name="quality-reviewer", description="Review reports for accuracy, clarity, and balance", system_prompt="""You are a quality assurance reviewer. Review the report critically: 1. Check factual accuracy against research 2. Assess clarity and readability 3. Verify balanced perspective (no bias) 4. Confirm proper citations 5. Suggest specific improvements Be thorough but constructive.""", tools=[] ) # 主代理 agent = create_deep_agent( model="anthropic:claude-sonnet-4-6", subagents=[ research_subagent, analysis_subagent, writer_subagent, critique_subagent ], system_prompt="""You are a policy research coordinator. When given a research topic: 1. First, delegate to policy-researcher to gather information 2. Then, pass research to policy-analyst for analysis 3. Then, pass analysis to report-writer for drafting 4. Finally, have quality-reviewer check the draft 5. Incorporate review feedback and deliver final report Each step should build on the previous work.""" ) # 使用 result = agent.invoke({ "messages": [{ "role": "user", "content": "Research the EU AI Act's requirements for high-risk AI systems and its global impact" }] })
GENERAL_PURPOSE_SUBAGENT: SubAgent = { "name": "general-purpose", "description": "General-purpose agent for researching complex questions...", "system_prompt": "In order to complete the objective...", }
粤ICP备14082021号
免费POC，
零成本试错
