LangChain’s prefab agent for when wiring feels like a chore

What it does

deepagents is a TypeScript agent harness built on LangGraph that ships with a pre-assembled toolkit—planning, filesystem access, sub-agent delegation, and context management. You call createDeepAgent and receive a compiled graph that can break down tasks, read and write files for working memory, and spawn isolated sub-agents. It is provider-agnostic across tool-calling chat models and publishes separate entrypoints for Node.js and browser runtimes.

The interesting bit

The project treats the filesystem as working memory, using file-based workflows to keep long tasks manageable instead of inventing exotic memory architectures. It also adopts an unusually honest security stance: it assumes you sandbox the tools because the LLM itself will not self-police.

Key highlights

• Ships with write_todos, read_file, write_file, edit_file, and sub-agent task tools out of the box
• Returns a compiled LangGraph graph, so streaming, checkpointers, and Studio integration work natively
• Provider-agnostic: works with any tool-calling chat model
• Environment-specific builds for Node.js and browser runtimes
• MIT-licensed and designed for incremental customization of tools, models, and prompts

Caveats

• The security model is explicitly “trust the LLM”; you must enforce boundaries at the tool or sandbox level, not via model behavior.

Verdict

Ideal for developers who want a production-ready LangGraph scaffold without boilerplate; less so for teams that need total architectural control or cannot tolerate unsandboxed LLM file access.