LangChain Releases OpenWiki: An Open-Source CLI to Automate Codebase Documentation for AI Agents

LangChain has released OpenWiki, an open-source TypeScript command-line interface (CLI) and AI agent designed to write and automatically maintain documentation for software repositories.

The tool addresses the common industry problem of “documentation rot”—where codebase documentation rapidly falls out of sync with actual code as developers push new commits. By scanning a repository’s file structure and git history, OpenWiki creates a dedicated, structured wiki directory designed to help AI coding agents understand the workspace they are operating in.

Shifting From Token-Heavy Prompt Files to Retrieval-Based Context

Traditionally, developers preparing a repository for AI coding assistants—such as Claude Code, Codex, or Gemini—rely on instruction files like CLAUDE.md or AGENTS.md. Because these models require structural context to avoid hallucinating file layouts or using outdated patterns, developers often append architectural guidelines directly to these markdown files.

However, as a codebase expands, these instruction files can balloon to hundreds of pages. Cramming extensive documentation into a single prompt inflates per-request API costs and degrades the model’s responsiveness, as the agent must parse the entire manual during every single run.

OpenWiki addresses this by shifting from context stuffing to a selective retrieval architecture. Instead of placing raw documentation directly into instruction files, the tool generates a dedicated openwiki/ directory at the project root containing structured Markdown files. It then automatically appends routing instructions to existing AGENTS.md or CLAUDE.md files, or creates them if they are missing. These instructions point coding agents to the openwiki/ folder, directing them to search and retrieve only the specific pages they need for a given task.

├── openwiki/
│   ├── quickstart.md       <-- Central index and repository map
│   └── [structured wiki pages]
├── CLAUDE.md               <-- Light instructions pointing agents to openwiki/
└── AGENTS.md               <-- Auto-injected routing prompts

How the Repository Analysis Works

OpenWiki operates as a global CLI distributed through the Node Package Manager (npm). When run for the first time on a local machine, the developer executes:

openwiki --init

This initialization command prompts the user to select an LLM provider and enter their API credentials. To protect against the risk of committing active API keys to public repositories, OpenWiki saves configuration details and secrets locally in a global, hidden file at ~/.openwiki/.env, keeping them isolated from project files.

Once configured, the underlying agent analyzes the target repository. It runs a sequence of Git commands to examine the directory structure, recent commit history, and any uncommitted diffs to generate a comprehensive wiki. The primary index file generated is quickstart.md, which serves as a central map guiding AI agents toward deeper technical files within the codebase.

OpenWiki is model-agnostic and supports several providers out of the box:

• OpenRouter (the default option, pre-configured with open models)
• Fireworks
• Baseten
• OpenAI
• Anthropic

Developers can also configure their own custom model IDs or hook the CLI into any OpenAI-compatible provider. For teams wishing to inspect the agent’s actions, the tool integrates with LangSmith—if a LangSmith API key is provided, OpenWiki will trace its execution runs to a project workspace for debugging.

Automating Updates via Scheduled Workflows

Generating codebase documentation once is a temporary fix; keeping it current is where technical debt typically accumulates. To automate maintenance, OpenWiki includes pre-configured workflows for continuous integration (CI) environments, including GitHub Actions and GitLab CI.

When triggered in a CI pipeline, the runner executes a non-interactive, headless command, pulling the necessary API credentials from secure repository environment variables rather than local storage.

The action executes the following update pipeline:

1. Change Detection: OpenWiki runs with an update flag (openwiki --update).
2. Git Diff Audit: The agent identifies which commits have been merged since the last documentation run.
3. Selective Rewrite: Using git diffs, the agent modifies only the wiki pages affected by the code changes.
4. Human Review Handoff: If changes are detected, the tool bundles the updated files and automatically submits a pull request (PR) or merge request to the repository.

By submitting changes as a PR rather than committing directly to the main branch, OpenWiki ensures a human developer remains in the loop to review and approve the AI-generated documentation before it is accepted as the new source of truth for downstream agents.

The project is fully open-source and licensed under the MIT License.