Deep Agents Documentation (LangChain)
by LangChain
The official docs for LangChain's batteries-included agent harness: planning, filesystem context, subagents, and memory.
Overview
Launched in March 2026, deepagents is described in its docs as "the easiest way to start building agents and applications that are powered by LLMs — with built-in capabilities for task planning, file systems for context management, subagent-spawning, and long-term memory." It is a standalone library built on LangChain's core agent building blocks and running on the LangGraph runtime for durable execution, positioned one level above create_agent: more opinionated, with the harness bundled in. The documentation is organized around four capability areas — execution environment (tools, a virtual filesystem with permission controls, sandboxed code execution, event streaming), context management (skills for domain knowledge, memory for persistent preferences, automatic summarization and offloading, prompt caching), delegation (task planning and subagent spawning with isolated context windows for parallel work), and steering (human-in-the-loop interrupts and filesystem permissions). It directly targets the practical failure modes of production agents: context windows filling on long tasks, multi-step work that needs decomposition, knowledge that must persist across sessions, and irreversible actions that need human approval. Teams wanting lower-level control are pointed back to create_agent or raw LangGraph workflows; deepagents also pairs with LangSmith for tracing and evaluation.
At a Glance
- Topic
- Frameworks
- Level
- Intermediate
- Format
- Documentation
- Cost
- Free
- Duration
- Reference documentation, self-paced
- Provider
- LangChain
- Hands-on
- Yes — code/exercises
- Certificate
- None
What You’ll Learn
- ✓Set up a deep agent with planning, virtual filesystem, subagents, and memory enabled by default
- ✓Manage long-running context with automatic summarization, offloading, and prompt caching
- ✓Delegate parallel work to subagents with isolated context windows
- ✓Add human-in-the-loop interrupts and filesystem permissions to gate risky actions
- ✓Run code safely via sandboxes and interpreters inside an agent's execution environment
- ✓Decide when to use deepagents versus create_agent or a raw LangGraph workflow
Highlights
- •Agent harness with production defaults instead of a bare agent loop
- •Built on the LangGraph runtime for durable, resumable execution
- •First-class Skills and long-term memory for context engineering
- •Open source, with LangSmith integration for tracing and evals
Who It’s For
Best For
- ✓Engineers building long-running or deep-research style agents
- ✓Teams already on LangChain/LangGraph wanting a higher-level harness
- ✓Developers who need subagent delegation and human-in-the-loop approval out of the box
Prerequisites
- •Python and familiarity with LLM tool calling
- •Basic understanding of LangChain or LangGraph agent concepts
FAQ
What is Deep Agents Documentation (LangChain)?
Official documentation for deepagents, LangChain's open-source agent harness for long-running, multi-step tasks. It is for engineers who want planning, context management, subagent delegation, and long-term memory as defaults rather than as hand-rolled scaffolding around a bare agent loop.
Is Deep Agents Documentation (LangChain) free?
Deep Agents Documentation (LangChain) is free to access.
What level is Deep Agents Documentation (LangChain) for?
Deep Agents Documentation (LangChain) is aimed at a intermediate audience. Recommended background: Python and familiarity with LLM tool calling, Basic understanding of LangChain or LangGraph agent concepts.
How long does Deep Agents Documentation (LangChain) take?
Expect roughly Reference documentation, self-paced. Most learners work through it at their own pace.
What will I learn from Deep Agents Documentation (LangChain)?
You'll learn: Set up a deep agent with planning, virtual filesystem, subagents, and memory enabled by default; Manage long-running context with automatic summarization, offloading, and prompt caching; Delegate parallel work to subagents with isolated context windows; Add human-in-the-loop interrupts and filesystem permissions to gate risky actions; Run code safely via sandboxes and interpreters inside an agent's execution environment; Decide when to use deepagents versus create_agent or a raw LangGraph workflow.