OpenWiki’s code init prompt just got a major upgrade! It now generates higher quality wikis that cover more of your codebase, leading to better eval scores and more efficient runs. Try it out locally today: https://github.com/langchain-ai/openwiki…

langchain-ai/openwiki

Source: https://github.com/langchain-ai/openwiki

The self-maintaining wiki. Built for agents, explored by humans.

OpenWiki is a CLI that writes and maintains a wiki for your codebase or your personal knowledge. An agent reads your sources, synthesizes a linked Markdown wiki you own, and keeps it current on every change. It is built for agents to read as memory, and it ships an interactive visualizer for humans to explore.

OpenWiki gives you:

• Agent-written docs that stay accurate, generated by a Deep Agents documentation agent.
• Two modes: a code wiki for a repository, or a personal wiki for your own knowledge.
• Twelve model providers out of the box, from OpenAI and Anthropic to Bedrock, Gemini, and any OpenAI-compatible gateway.
• Built-in connectors for Notion, Slack, Gmail, X, Web Search, Hacker News, and local git repositories.
• An interactive visualizer that turns any wiki into a live, explorable node graph.
• Self-updating through GitHub Actions, GitLab CI, or Bitbucket Pipelines.
• Open Knowledge Format (OKF v0.1) output with validated Mermaid diagrams.

🎉 What’s new

• Interactive visualizer: turn any wiki into a live, explorable node graph with a side-by-side Markdown reader.
• .openwikiignore: keep generated, private, or irrelevant paths out of doc runs with familiar gitignore-style rules.
• Multilingual wikis: generate docs in another language with --language <locale>, while code and identifiers stay canonical.
• LangSmith connector: pull recent LangSmith traces (tool calls, outcomes, latency) into a code wiki.
• GitHub Copilot provider: reuse an existing Copilot subscription for inference, no separate API key required.

Quick start

Install the CLI:

npm install -g openwiki

Generate a wiki for the current repository. The first run walks you through picking a provider, key, and model, then writes docs to openwiki/:

openwiki --init

Keep it current automatically by adding a scheduled CI job that opens a docs PR on every change:

• GitHub Actions: copy openwiki-update.yml into .github/workflows/openwiki-update.yml.
• GitLab CI: copy openwiki-update.gitlab-ci.yml into .gitlab-ci.yml or include it from your pipeline.
• Bitbucket Pipelines: copy openwiki-update.bitbucket-pipelines.yml into bitbucket-pipelines.yml, then schedule the openwiki-update pipeline.

On Windows, install with a Node.js package manager (npm install -g openwiki or pnpm add -g openwiki). Installing with bun can fall back to compiling the better-sqlite3 native dependency, which needs Visual Studio Build Tools with the Desktop development with C++ workload.

Two modes

OpenWiki runs in one of two modes. Bare openwiki, openwiki --init, and openwiki --update default to code mode; add the personal positional (or --mode personal) for the personal brain.

By default the CLI stays open after a run so you can send follow-up messages. Add -p / --print for a one-shot, non-interactive run that prints the final output and exits. --init and --update auto-exit on success in an interactive terminal, so the same command works one-shot or interactively.

Explore your wiki

Turn any wiki into an interactive node graph with a live, side-by-side Markdown reader:

openwiki visualize

This serves ./openwiki on a local loopback address (127.0.0.1, never exposed on the network) and opens your browser to the graph. Edits to the wiki files are picked up automatically while the server runs. Pass a path to visualize a different directory, --port <port> to choose the port (it increments on conflict; default 4321), and --no-open to leave the browser alone:

openwiki visualize openwiki --port 4400 --no-open

The page loads its graph, Markdown, and diagram libraries from a public CDN, so an internet connection is required even though the server itself is local. Press Ctrl-C to stop it.

Connect your sources

In personal mode, OpenWiki ingests knowledge from the tools you already use, synthesizing them into your local wiki. First-run onboarding offers setup for local git repositories, Notion, Gmail, X/Twitter, Web Search, and Hacker News.

During an ingestion run, deterministic connector tools write raw data and manifests under ~/.openwiki/connectors/<connector>/raw/, then source-specific agent runs synthesize the wiki under ~/.openwiki/wiki/. You can configure the same connector more than once (for example one Web Search source for AI research and another for NBA news); OpenWiki stores them as separate instances like web-search-1 and web-search-2.

openwiki auth notion        # run a local browser OAuth flow for a provider
openwiki ingest all         # run every configured source
openwiki ingest web-search  # run one connector's sources

LangSmith connector (code mode)

The connectors above feed a personal wiki. The LangSmith connector instead enriches a code wiki: it pulls recent LangSmith traces (tool calls, outcomes, and latency) for the projects you choose through the official LangSmith SDK, so a repository’s docs reflect how its code actually behaves at runtime, not just what the source says.

Configure it during openwiki --init in code mode. From the source menu, add LangSmith, pick your workspace region (US or EU), and list the projects to document. OpenWiki writes a committed openwiki/.langsmith.json that names the workspaces and projects (never the key itself), so every teammate and CI run documents the same set. The API key is read from the environment:

OPENWIKI_LANGSMITH_API_KEY="<your-langsmith-key>"

Locally the setup wizard saves this to ~/.openwiki/.env. In CI, set it as a repository secret and export it for the run.

A LangSmith key is workspace- and region-bound. To document projects across more than one workspace, add an entry per workspace, each with its own key named OPENWIKI_LANGSMITH_API_KEY_2, OPENWIKI_LANGSMITH_API_KEY_3, and so on. The connector only talks to the official US (api.smith.langchain.com) and EU (eu.api.smith.langchain.com) hosts.

How it stays yours

Everything OpenWiki writes is plain Markdown you own and version alongside your code.

• Agents read it as memory. On each code run, OpenWiki maintains an AGENTS.md and CLAUDE.md at the repo root that point your coding agent at the wiki. It only rewrites its own <!-- OPENWIKI:START -->…<!-- OPENWIKI:END --> block and leaves the rest of each file untouched.
• You set the brief. Repository-specific instructions live in openwiki/INSTRUCTIONS.md, a user-authored file OpenWiki reads for scope and priorities but never rewrites during normal runs.
• No-op runs are free. After a run, OpenWiki snapshots the openwiki/ directory and only records new metadata when something actually changed, so scheduled workflows never churn.
• Local, private config. Provider choice, keys, and optional LangSmith tracing are saved to ~/.openwiki/.env on your machine.

Open Knowledge Format

OpenWiki emits Google Open Knowledge Format (OKF) v0.1 bundles in both modes, so your wiki is portable to any OKF-aware tool.

• Every concept document carries YAML front matter with a non-empty type; all other standard fields are optional.
• Standard Markdown links between concept documents express their relationships.
• index.md and log.md are reserved documents rather than concepts. The root index declares okf_version: "0.1".
• Valid timestamp values and producer-defined extension fields are preserved across updates and migrations.

Diagrams

OpenWiki embeds Mermaid diagrams wherever they make a concept clearer than prose: sequence diagrams for runtime flows, ER diagrams for data models, state diagrams for lifecycles, and flowcharts for control flow. Diagrams are grounded in the inspected source, added where they add signal, and kept in sync on --update. No configuration is required.

After each run, OpenWiki validates every mermaid fence. A diagram that fails validation is converted in place to a plain text fence with a short comment explaining why, so it degrades to readable text instead of a broken block. The next --update finds that comment and repairs the diagram, so quality recovers over successive runs.

By default OpenWiki runs a lightweight, zero-dependency check that catches common breakages. For authoritative validation that matches exactly what GitHub renders, install the Mermaid parser wherever you run OpenWiki (for example in your scheduled workflow), and no broken diagram will ship:

npm install mermaid jsdom

Model providers

The onboarding default is OpenAI with gpt-5.6-terra. Every provider includes preset model options plus support for custom model IDs, and stores its credentials in ~/.openwiki/.env.

OPENWIKI_PROVIDER="copilot"
OPENWIKI_MODEL_ID="gpt-5.5"

OPENWIKI_PROVIDER=bedrock
BEDROCK_AWS_ACCESS_KEY_ID=your-access-key-id
BEDROCK_AWS_SECRET_ACCESS_KEY=your-secret-access-key
BEDROCK_AWS_REGION=us-east-1
OPENWIKI_MODEL_ID=anthropic.claude-sonnet-5

OPENWIKI_PROVIDER=gemini
GEMINI_API_KEY=your-ai-studio-key

OPENWIKI_PROVIDER=gemini-enterprise
GOOGLE_CLOUD_PROJECT=your-gcp-project
GOOGLE_CLOUD_LOCATION=global   # optional, defaults to global

OPENWIKI_PROVIDER=openai-chatgpt openwiki code --init
# or
OPENWIKI_PROVIDER=openai-chatgpt openwiki personal --init

# Hosted gateway (for example Requesty, which fronts many upstream providers)
OPENWIKI_PROVIDER=openai-compatible
OPENAI_COMPATIBLE_API_KEY=your-gateway-key
OPENAI_COMPATIBLE_BASE_URL=https://router.requesty.ai/v1
OPENWIKI_MODEL_ID=openai/gpt-5.5

# Ollama, after `ollama serve` and `ollama pull llama3.2`
OPENWIKI_PROVIDER=openai-compatible
OPENAI_COMPATIBLE_API_KEY=ollama
OPENAI_COMPATIBLE_BASE_URL=http://localhost:11434/v1
OPENWIKI_MODEL_ID=llama3.2

# LM Studio, after starting the local server from the Developer tab
OPENWIKI_PROVIDER=openai-compatible
OPENAI_COMPATIBLE_API_KEY=lm-studio
OPENAI_COMPATIBLE_BASE_URL=http://localhost:1234/v1
OPENWIKI_MODEL_ID=your-loaded-model-id

OPENWIKI_PROVIDER=anthropic
ANTHROPIC_API_KEY=your-key
ANTHROPIC_BASE_URL=https://your-gateway.example.com/anthropic

OPENWIKI_PROVIDER=openrouter
OPENROUTER_API_KEY=your-key
OPENWIKI_OPENROUTER_PROVIDER_ONLY=Novita

If there is an inference provider or model you would like to see added, please open a PR.

Ignoring paths

Create a .openwikiignore file in the repository root to keep generated docs from reading or describing private, generated, or irrelevant paths. The syntax supports comments, blank lines, * and ** globs, directory rules, and ! negation:

secrets/
*.log
!logs/keep.log

When .openwikiignore has active rules, OpenWiki filters filesystem discovery and restricts shell execute so ignored paths stay out of the run. This is a read boundary: ignored paths are never read, scanned, or reproduced in the docs. It does not guarantee a topic is never mentioned, since the agent may still infer an ignored area from other allowed evidence such as tests, the README, or commit messages.

Command reference

openwiki                         # interactive chat, code mode, current repo
openwiki personal                # interactive chat, personal brain
openwiki "generate docs"         # start with an initial request
openwiki -p "what can you do?"   # one-shot, print, and exit
openwiki --init                  # initialize code docs (personal: openwiki personal --init)
openwiki --update                # update code docs (personal: openwiki personal --update)
openwiki visualize               # interactive graph + live reader
openwiki auth <provider>         # authenticate a connector (slack, gmail, x, notion)
openwiki ingest <source>         # run connector ingestion (all, or a connector/instance)
openwiki --help                  # full help

In chat, /api-key updates the current provider key and /langsmith-key updates or clears LangSmith tracing credentials, both with masked prompts.

Telemetry

OpenWiki collects anonymous, aggregate usage data to understand how the tool is used and improve it. Telemetry is on by default and easy to turn off.

Collected on a single openwiki_run event, keyed by a random install ID in ~/.openwiki/install-id: the command (init / update), the outcome (success / failure / no-op) with a coarse error category on failure (never the message), and at setup only, the brain mode, model provider, and configured connector names.

Never collected: file contents, repository data or names, credentials, prompts, model output, connector payloads, error messages, file paths, URLs, model IDs, run duration, or your IP address. Interactive chat, auth, and ingest are not recorded. Scheduled/CI runs are tagged as anonymous reliability data under a shared CI identifier and never counted as installs.

Opt out with either environment variable, or add the first line to ~/.openwiki/.env to disable permanently:

export OPENWIKI_TELEMETRY_DISABLED=1
export DO_NOT_TRACK=1   # cross-tool standard

To see exactly what a run would send, add --telemetry-file=<path> to any run.

Contributing

Contributions are welcome. Please read CONTRIBUTING.md before opening a PR. We intentionally keep PRs tightly scoped to one change each, and PRs that bundle unrelated changes may be closed with a request to split them.

License

MIT

Brace (@BraceSproul): I just rewrote OpenWiki’s entire prompt in v0.3 to generate more detailed & accurate wikis. Eval results:

• 📈 28.57% success increase (35% to 45% @ n=2)
• 🪙 14% fewer tokens
• 🧰 26% fewer tool calls per successful task

And as you can clearly see from the graphs, the wikis