LangChain's new open-source agent builds wikis for your code

OpenWiki is an open-source CLI agent from LangChain that automatically writes documentation for your codebase and keeps it updated as your code evolves, giving coding agents like Claude Code or Cursor the exact context they need.

AI coding assistants write significantly better code when they understand your repo's layout, where core logic lives, and what patterns to follow. But most developers try to solve this by dumping everything into a CLAUDE.md or AGENTS.md file. When a codebase grows to hundreds of files, cramming all that context into a single prompt blows up token budgets and confuses the model.

OpenWiki solves this with a structured approach inspired by Karpathy's LLM Wiki concept, DeepWiki, and AutoWiki. Instead of bloating your instruction files, OpenWiki generates a modular wiki across multiple markdown pages. Then, it drops a short reference into your CLAUDE.md or AGENTS.md, telling your coding agent where the wiki lives and how to look up specific modules on demand.

Getting started in minutes

Setting up OpenWiki is straightforward through npm.

You install the CLI globally by running npm install -g openwiki, navigate to your project directory, and run openwiki --init. The command prompts you for a model provider and an API key, then scans the repository and generates your initial documentation.

OpenWiki works with both open-weight and proprietary providers, including OpenRouter, Fireworks, Baseten, OpenAI, and Anthropic. By default, it runs on OpenRouter using an open model.

Because the tool is built on LangChain's DeepAgents, it also includes native LangSmith tracing. If you supply a LangSmith API key, you can inspect the agent's exact reasoning steps as it analyzes your codebase and writes docs.

The community has already started porting the pattern elsewhere. An open-source extension named pi-openwiki adapts the agent for the Pi terminal harness, letting developers run wiki generation and repo queries directly through slash commands.

How it stays in sync with git

Generating docs once is neat, but static docs rot quickly in active repos. OpenWiki's real leverage comes from automated incremental syncs.

The tool ships with a prebuilt GitHub Action meant to run on a schedule, such as once a day. When triggered with the update flag, OpenWiki analyzes git diffs for any commits merged since the previous run, pinpoints modified logic, and selectively edits the relevant wiki pages.

Everything happens silently in your CI pipeline. Developers do not need to pause their workflow to update markdown files manually, and your coding agent always pulls current architectural context whenever you kick off a task.

You can clone the source code on GitHub at langchain-ai/openwiki or install it directly via npm.

Why it matters

If you work on a multi-module repository and rely on AI coding tools, OpenWiki is worth testing right away. It treats codebase documentation not as a chore for humans, but as an external hard drive for AI agents that frees up your instruction files.

This marks a practical shift in agent architecture. Stuffing thousands of lines into system prompts was always a crude workaround; offloading durable knowledge into structured, git-aware wikis is a far cleaner pattern. LangChain nailed the ergonomics here by relying on standard git workflows rather than forcing teams into proprietary platforms.