OpenWiki
The self-maintaining wiki. Built for agents, explored by humans.
About OpenWiki
OpenWiki is a CLI from LangChain that writes and maintains a wiki for a codebase. A documentation agent reads your sources, synthesises a linked Markdown wiki that lives in your repository and belongs to you, and updates it as the code changes. It is built to be read two ways: by coding agents, which use it as memory instead of re-deriving how a system works on every session, and by humans, through an interactive visualiser that turns the wiki into an explorable node graph.
Generated documentation has one well-known failure mode, and it is not that the first draft is bad. It is that the draft is decent, the code moves, and six weeks later the docs confidently describe a system that no longer exists - which is worse than having no docs, because people believe them. OpenWiki's answer is Grounded Claims: material facts in a code wiki carry versioned evidence pointing at the source they came from, so when that evidence changes or disappears the tool knows precisely which propositions need to be confirmed, rewritten or retired. Whether that holds up in practice is the thing to evaluate before you trust it, and it is the reason to read the first generated wiki carefully rather than merging it. The project is young and moving: it is at an early 0.x version, and the repository generation flow was recently rebuilt into a resumable page-job pipeline, so expect the shape of the output to keep changing for a while yet.
Install openwiki from npm and run it against a repository in code mode, or against your own notes and sources in personal mode. The agent plans the wiki, then works page by page through a durable ordered queue: it submits a plan, generates each page with a focused worker, persists the Claims and source checkpoints for that page, and finishes. Because the queue is durable and checkpointed, a run that dies partway through does not start over, and in ephemeral CI it can open a pull request with partial progress rather than losing the work.
Keeping it current is a scheduled job rather than a habit you have to maintain. OpenWiki self-updates through GitHub Actions, GitLab CI or Bitbucket Pipelines, so the wiki is regenerated and re-grounded as commits land. For reading, the visualiser renders any wiki as a live node graph you can explore.
- •Grounded Claims - Material facts are tracked back to versioned source evidence, so when the evidence changes OpenWiki can surface exactly which statements are now suspect instead of quietly leaving them wrong
- •Two Modes - A
codewiki for a repository, or apersonalwiki built from your own knowledge and sources - •Thirteen Model Providers - OpenAI, Anthropic, Bedrock, Gemini and any OpenAI-compatible gateway, so the documentation agent is not tied to one vendor
- •Coding Agent Integrations - Runs inside Codex, Claude Code, OpenCode or Cursor using the host agent's authenticated model and native repository tools, while OpenWiki manages the documentation lifecycle
- •Nine Built-In Connectors - Custom MCP, Notion, Slack, Gmail, X, Web Search, Hacker News, LangSmith and local git repositories as sources
- •Resumable Page-Job Architecture - A durable ordered page queue with per-page workers, strict Claims persistence and per-page source checkpoints, plus partial-progress pull requests for ephemeral CI
- •Interactive Visualiser - Any wiki renders as a live, explorable node graph for humans
- •Open Knowledge Format Output - Emits OKF v0.2 with validated Mermaid diagrams, so the wiki is a portable artifact rather than a proprietary format
Teams whose codebase is large enough that onboarding is expensive and whose documentation has already fallen behind, and anyone running coding agents against a big repository who is tired of the agent rediscovering the same architecture every session - the wiki is explicitly designed as agent memory. The CI integration makes it a fit for a team that wants documentation maintained as a pipeline step rather than as somebody's neglected side task.
Be careful in two cases. If your documentation carries compliance or contractual weight, a generated wiki needs review before anyone relies on it, and Grounded Claims is a mechanism for catching drift rather than a guarantee of correctness. And if you want a supported product, note that this has no website, no hosted tier and no support channel beyond GitHub issues, and it is built on LangChain's Deep Agents, so it inherits that stack's pace of change.
Pricing
No pricing page found
- Free (MIT, self-hosted CLI)$0/mo
Checked 2026-09-30














