"""Deep Agents come with planning, filesystem, and subagents.""" from collections.abc import Callable, Sequence from typing import Any from langchain.agents import create_agent from langchain.agents.middleware import HumanInTheLoopMiddleware, InterruptOnConfig, TodoListMiddleware from langchain.agents.middleware.types import AgentMiddleware from langchain.agents.structured_output import ResponseFormat from langchain.chat_models import init_chat_model from langchain_anthropic import ChatAnthropic from langchain_anthropic.middleware import AnthropicPromptCachingMiddleware from langchain_core.language_models import BaseChatModel from langchain_core.messages import SystemMessage from langchain_core.tools import BaseTool from langgraph.cache.base import BaseCache from langgraph.graph.state import CompiledStateGraph from langgraph.store.base import BaseStore from langgraph.types import Checkpointer from deepagents.backends import StateBackend from deepagents.backends.protocol import BackendFactory, BackendProtocol from deepagents.middleware.filesystem import FilesystemMiddleware from deepagents.middleware.memory import MemoryMiddleware from deepagents.middleware.patch_tool_calls import PatchToolCallsMiddleware from deepagents.middleware.skills import SkillsMiddleware from deepagents.middleware.subagents import ( GENERAL_PURPOSE_SUBAGENT, CompiledSubAgent, SubAgent, SubAgentMiddleware, ) from deepagents.middleware.summarization import create_summarization_middleware BASE_AGENT_PROMPT = """You are a Deep Agent, an AI assistant that helps users accomplish tasks using tools. You respond with text and tool calls. The user can see your responses and tool outputs in real time. ## Core Behavior - Be concise and direct. Don't over-explain unless asked. - NEVER add unnecessary preamble (\"Sure!\", \"Great question!\", \"I'll now...\"). - Don't say \"I'll now do X\" — just do it. - If the request is ambiguous, ask questions before acting. - If asked how to approach something, explain first, then act. ## Professional Objectivity - Prioritize accuracy over validating the user's beliefs - Disagree respectfully when the user is incorrect - Avoid unnecessary superlatives, praise, or emotional validation ## Doing Tasks When the user asks you to do something: 1. **Understand first** — read relevant files, check existing patterns. Quick but thorough — gather enough evidence to start, then iterate. 2. **Act** — implement the solution. Work quickly but accurately. 3. **Verify** — check your work against what was asked, not against your own output. Your first attempt is rarely correct — iterate. Keep working until the task is fully complete. Don't stop partway and explain what you would do — just do it. Only yield back to the user when the task is done or you're genuinely blocked. **When things go wrong:** - If something fails repeatedly, stop and analyze *why* — don't keep retrying the same approach. - If you're blocked, tell the user what's wrong and ask for guidance. ## Progress Updates For longer tasks, provide brief progress updates at reasonable intervals — a concise sentence recapping what you've done and what's next.""" # noqa: E501 def get_default_model() -> ChatAnthropic: """Get the default model for deep agents. Returns: `ChatAnthropic` instance configured with Claude Sonnet 4.6. """ return ChatAnthropic( model_name="claude-sonnet-4-6", ) def resolve_model(model: str | BaseChatModel) -> BaseChatModel: """Resolve a model string to a `BaseChatModel` instance. If `model` is already a `BaseChatModel`, returns it unchanged. String models are resolved via `init_chat_model`, with OpenAI models defaulting to the Responses API. See the `create_deep_agent` docstring for details on how to customize this behavior. Args: model: Model name string or pre-configured model instance. Returns: Resolved `BaseChatModel` instance. """ if isinstance(model, BaseChatModel): return model if model.startswith("openai:"): # Use Responses API by default. To use chat completions, use # `model=init_chat_model("openai:...")` # To disable data retention with the Responses API, use # `model=init_chat_model("openai:...", use_responses_api=True, store=False, include=["reasoning.encrypted_content"])` return init_chat_model(model, use_responses_api=True) return init_chat_model(model) def create_deep_agent( # noqa: C901, PLR0912 # Complex graph assembly logic with many conditional branches model: str | BaseChatModel | None = None, tools: Sequence[BaseTool | Callable | dict[str, Any]] | None = None, *, system_prompt: str | SystemMessage | None = None, middleware: Sequence[AgentMiddleware] = (), subagents: list[SubAgent | CompiledSubAgent] | None = None, skills: list[str] | None = None, memory: list[str] | None = None, response_format: ResponseFormat | None = None, context_schema: type[Any] | None = None, checkpointer: Checkpointer | None = None, store: BaseStore | None = None, backend: BackendProtocol | BackendFactory | None = None, interrupt_on: dict[str, bool | InterruptOnConfig] | None = None, debug: bool = False, name: str | None = None, cache: BaseCache | None = None, ) -> CompiledStateGraph: """Create a deep agent. !!! warning "Deep agents require a LLM that supports tool calling!" By default, this agent has access to the following tools: - `write_todos`: manage a todo list - `ls`, `read_file`, `write_file`, `edit_file`, `glob`, `grep`: file operations - `execute`: run shell commands - `task`: call subagents The `execute` tool allows running shell commands if the backend implements `SandboxBackendProtocol`. For non-sandbox backends, the `execute` tool will return an error message. Args: model: The model to use. Defaults to `claude-sonnet-4-6`. Use the `provider:model` format (e.g., `openai:gpt-5`) to quickly switch between models. If an `openai:` model is used, the agent will use the OpenAI Responses API by default. To use OpenAI chat completions instead, initialize the model with `init_chat_model("openai:...", use_responses_api=False)` and pass the initialized model instance here. To disable data retention with the Responses API, use `init_chat_model("openai:...", use_responses_api=True, store=False, include=["reasoning.encrypted_content"])` and pass the initialized model instance here. tools: The tools the agent should have access to. In addition to custom tools you provide, deep agents include built-in tools for planning, file management, and subagent spawning. system_prompt: Custom system instructions to prepend before the base deep agent prompt. If a string, it's concatenated with the base prompt. middleware: Additional middleware to apply after the standard middleware stack (`TodoListMiddleware`, `FilesystemMiddleware`, `SubAgentMiddleware`, `SummarizationMiddleware`, `AnthropicPromptCachingMiddleware`, `PatchToolCallsMiddleware`). subagents: The subagents to use. Each subagent should be a `dict` with the following keys: - `name` - `description` (used by the main agent to decide whether to call the sub agent) - `system_prompt` (used as the system prompt in the subagent) - (optional) `tools` - (optional) `model` (either a `LanguageModelLike` instance or `dict` settings) - (optional) `middleware` (list of `AgentMiddleware`) skills: Optional list of skill source paths (e.g., `["/skills/user/", "/skills/project/"]`). Paths must be specified using POSIX conventions (forward slashes) and are relative to the backend's root. When using `StateBackend` (default), provide skill files via `invoke(files={...})`. With `FilesystemBackend`, skills are loaded from disk relative to the backend's `root_dir`. Later sources override earlier ones for skills with the same name (last one wins). memory: Optional list of memory file paths (`AGENTS.md` files) to load (e.g., `["/memory/AGENTS.md"]`). Display names are automatically derived from paths. Memory is loaded at agent startup and added into the system prompt. response_format: A structured output response format to use for the agent. context_schema: The schema of the deep agent. checkpointer: Optional `Checkpointer` for persisting agent state between runs. store: Optional store for persistent storage (required if backend uses `StoreBackend`). backend: Optional backend for file storage and execution. Pass either a `Backend` instance or a callable factory like `lambda rt: StateBackend(rt)`. For execution support, use a backend that implements `SandboxBackendProtocol`. interrupt_on: Mapping of tool names to interrupt configs. Pass to pause agent execution at specified tool calls for human approval or modification. Example: `interrupt_on={"edit_file": True}` pauses before every edit. debug: Whether to enable debug mode. Passed through to `create_agent`. name: The name of the agent. Passed through to `create_agent`. cache: The cache to use for the agent. Passed through to `create_agent`. Returns: A configured deep agent. """ model = get_default_model() if model is None else resolve_model(model) backend = backend if backend is not None else (StateBackend) # Build general-purpose subagent with default middleware stack gp_middleware: list[AgentMiddleware[Any, Any, Any]] = [ TodoListMiddleware(), FilesystemMiddleware(backend=backend), create_summarization_middleware(model, backend), AnthropicPromptCachingMiddleware(unsupported_model_behavior="ignore"), PatchToolCallsMiddleware(), ] if skills is not None: gp_middleware.append(SkillsMiddleware(backend=backend, sources=skills)) if interrupt_on is not None: gp_middleware.append(HumanInTheLoopMiddleware(interrupt_on=interrupt_on)) general_purpose_spec: SubAgent = { # ty: ignore[missing-typed-dict-key] **GENERAL_PURPOSE_SUBAGENT, "model": model, "tools": tools or [], "middleware": gp_middleware, } # Process user-provided subagents to fill in defaults for model, tools, and middleware processed_subagents: list[SubAgent | CompiledSubAgent] = [] for spec in subagents or []: if "runnable" in spec: # CompiledSubAgent - use as-is processed_subagents.append(spec) else: # SubAgent - fill in defaults and prepend base middleware subagent_model = spec.get("model", model) subagent_model = resolve_model(subagent_model) # Build middleware: base stack + skills (if specified) + user's middleware subagent_middleware: list[AgentMiddleware[Any, Any, Any]] = [ TodoListMiddleware(), FilesystemMiddleware(backend=backend), create_summarization_middleware(subagent_model, backend), AnthropicPromptCachingMiddleware(unsupported_model_behavior="ignore"), PatchToolCallsMiddleware(), ] subagent_skills = spec.get("skills") if subagent_skills: subagent_middleware.append(SkillsMiddleware(backend=backend, sources=subagent_skills)) subagent_middleware.extend(spec.get("middleware", [])) processed_spec: SubAgent = { # ty: ignore[missing-typed-dict-key] **spec, "model": subagent_model, "tools": spec.get("tools", tools or []), "middleware": subagent_middleware, } processed_subagents.append(processed_spec) # Combine GP with processed user-provided subagents all_subagents: list[SubAgent | CompiledSubAgent] = [general_purpose_spec, *processed_subagents] # Build main agent middleware stack deepagent_middleware: list[AgentMiddleware[Any, Any, Any]] = [ TodoListMiddleware(), ] if memory is not None: deepagent_middleware.append(MemoryMiddleware(backend=backend, sources=memory)) if skills is not None: deepagent_middleware.append(SkillsMiddleware(backend=backend, sources=skills)) deepagent_middleware.extend( [ FilesystemMiddleware(backend=backend), SubAgentMiddleware( backend=backend, subagents=all_subagents, ), create_summarization_middleware(model, backend), AnthropicPromptCachingMiddleware(unsupported_model_behavior="ignore"), PatchToolCallsMiddleware(), ] ) if middleware: deepagent_middleware.extend(middleware) if interrupt_on is not None: deepagent_middleware.append(HumanInTheLoopMiddleware(interrupt_on=interrupt_on)) # Combine system_prompt with BASE_AGENT_PROMPT if system_prompt is None: final_system_prompt: str | SystemMessage = BASE_AGENT_PROMPT elif isinstance(system_prompt, SystemMessage): final_system_prompt = SystemMessage(content_blocks=[*system_prompt.content_blocks, {"type": "text", "text": f"\n\n{BASE_AGENT_PROMPT}"}]) else: # String: simple concatenation final_system_prompt = system_prompt + "\n\n" + BASE_AGENT_PROMPT return create_agent( model, system_prompt=final_system_prompt, tools=tools, middleware=deepagent_middleware, response_format=response_format, context_schema=context_schema, checkpointer=checkpointer, store=store, debug=debug, name=name, cache=cache, ).with_config({"recursion_limit": 1000})