Open sourcePython · ≈31k stars · MIT
DeepAgents
The batteries-included agent harness.
DeepAgents packages the patterns that make research and long-horizon agents work — planning, sub-agents, file-system memory and a tool-calling loop — into a small library that runs on any LangGraph backend. The 'just give me the harness' option from the LangChain team.
Pros & cons
Each point cites its sources and says what kind of evidence backs it. A second reviewer tried to refute every claim before it was kept.
Pros
- DocsBatteries-included harness: sub-agents with isolated context, pluggable filesystem backends, thread summarization, human-in-the-loop approvals and skills out of the box.[1]
- DocsModel-agnostic: any LangChain chat model, including self-hosted Ollama, vLLM or llama.cpp.[1]
- DocsBuilt on LangGraph, so it inherits streaming, persistence and checkpointing, and has a JS/TS sibling (deepagents.js).[1]
- MetadataFrequent releases (0.7.32 on 2026-10-23) and about 31k stars.[1]
Cons
- DocsUses a 'trust the LLM' security model: the agent can do anything its tools allow, so you must enforce boundaries in the sandbox or tools yourself.[1]
- User reportsTied to the LangChain/LangGraph stack and pushes LangSmith for tracing and deployment; HN commenters call the 'deep agents' framing renamed agents-plus-tools.[1][2]
- User reportsSub-agent and approval edge cases have caused bugs: an infinite loop with a custom CompiledSubAgent, a sub-agent checkpointer problem, and model looping after a 'reject' decision (all fixed and closed).[1][2][3]
- Best for
- Python or JS teams already on LangGraph who want a Claude Code-style long-horizon agent without building planning and sub-agents themselves.
- Watch out for
- Security is delegated to your sandbox and tools, and observability nudges you toward paid LangSmith.
Assessed Oct 29, 2026 · adversarially reviewedSuggest a correction
Activity
- Last push
- · yesterday
- Latest release
- deepagents-acp==0.1.0
· yesterday - Last 90 days
- 20+ releases
- · yesterday
⚠ BREAKING CHANGES Requires agent-client-protocol=1.0.0, Git log since deepagents-acp==0.0.13 This commit history includes changes to this package. The most recent commit is at the top; older commits follow below. 9dca61f release(deepagents-acp): 0.1.0 (#7822) 0de4aac…
- · yesterday
Bug Fixes Correctly report unapproved MCP servers disabled via F2 as disabled, and report project MCP servers skipped due to missing approval (#7908, #7904). Show the source file in the project MCP trust prompt (#7903). Full changelog --- Internal maintainers: @darieldatoon,…
- · yesterday
Features Added Gemini 4 Neon and GPT-6.2 Astra and Sol to the model picker (#7861, #7831). Bug Fixes Changed OpenAI prompt_cache_key from per-thread to shared across threads within a local profile, improving prompt-cache reuse across sessions (#7839). Fixed Auto mode to resolve…
- · yesterday
Bug Fixes Fixed message-channel handling by using LangGraph's _messages_delta_reducer (#7819). Removed an incorrect claim that unread files cause an error (#7823). Removed redundant backend information from the execute tool description (#7862). Full changelog --- Special thanks…
- · yesterday
Features Ask structured clarifying questions in chat. (#7607) Inspect and guide background subagents. (#7609) Enable or disable configured MCP servers, and use skill-scoped MCP tools. (#7608, #7280) Add command hooks for the main agent and local subagents. (#7380, #7381) Filter…
Checked against GitHub on Nov 21, 2026. Release notes are shortened to plain text.
The team behind it
San Francisco, CA · Est. 2022
The de-facto standard toolkit for building LLM applications and agents.
LangChain makes the LangChain framework and LangGraph, the low-level orchestration standard for stateful, controllable agents, plus the LangSmith platform for tracing, evaluating and monitoring them in production.
~$75M raised
In research (1)
Oct 27, 2026 · The lowest-scoring harness for the frozen model (47.9%).
Neighbours on the workbench
TypeScript · ≈395k stars · MIT
openclaw/openclaw
The viral open-source personal AI assistant that really does things.
Maintained by OpenClaw Foundation
v2026.11.6releasedShell · ≈329k stars · MIT
mattpocock/skills
Small, composable agent skills for real engineering work.
Maintained by Matt Pocock (AI Hero)
MDX · ≈323k stars · MIT
obra/superpowers
An agentic skills framework and software methodology that works.
Maintained by Jesse Vincent (obra)
Connections
Relations from the Harness Hub knowledge graph, as charted in the State of Harness reports.
Maintained by · 1
- LangChainStartup
