OpenWiki is a CLI that writes and maintains agent wikis for codebases or purpose memory. It's built specifically for agents, can ingest local knowledge sources through built-in connectors or git repositories and synthesize them into a local wiki.
npm install -g openwikiOn Windows, prefer installing OpenWiki with Node.js package managers such as
npm or pnpm:
npm install -g openwiki
# or
pnpm add -g openwikibun install -g openwiki can fall back to compiling OpenWiki's better-sqlite3
checkpointing dependency. Before using that path, install Visual Studio Build
Tools with the Desktop development with C++ workload. Bun does not run lifecycle
scripts from installed packages by default, so it cannot display a package-level
warning before that native dependency build starts.
Initialize OpenWiki in code mode, configure your model and API key, then generate documentation:
openwiki --initOpenWiki has two modes:
- Personal mode builds a local personal brain wiki in
~/.openwiki/wikifrom configured sources like local repositories, Gmail, Notion, Web Search, Hacker News, and X/Twitter. - Code mode builds repository documentation in
openwiki/for the current codebase.
Bare openwiki --init and openwiki --update run in code mode. Use
openwiki personal --init or openwiki personal --update for the local
personal brain wiki.
Then to ensure your documentation stays up-to-date, add the CI workflow for your Git provider to automatically open a PR or merge request with documentation updates:
- GitHub Actions: copy openwiki-update.yml into
.github/workflows/openwiki-update.yml. - GitLab CI: copy openwiki-update.gitlab-ci.yml into
.gitlab-ci.ymlor include it from your existing GitLab pipeline. - Bitbucket Pipelines: copy openwiki-update.bitbucket-pipelines.yml into
bitbucket-pipelines.yml, then schedule theopenwiki-updatecustom pipeline from Repository settings > Pipelines > Schedules.
For repository documentation in GitHub Actions, use
openwiki code --update --print. You do not need to run --init in CI:
--update will create the initial openwiki/ docs if they do not exist yet, as
long as the workflow provides the required provider and model environment
variables.
Scheduled/CI runs send anonymous reliability telemetry. See Telemetry
for what is collected and how to turn it off (uncomment OPENWIKI_TELEMETRY_DISABLED
in the example workflow).
Start the interactive CLI in code mode for the current repository:
openwikiStart OpenWiki with an initial request:
openwiki "Please generate documentation for this repository"Start the interactive local personal brain instead:
openwiki personalRun a single command and exit:
openwiki -p "Summarize what you can do"Initialize OpenWiki:
openwiki --initInitialize the local personal brain wiki:
openwiki personal --initUpdate repository code documentation:
openwiki --updateUpdate the local personal brain wiki:
openwiki personal --updateRun an update that can ingest configured local connectors first:
openwiki personal --update "Refresh the wiki from configured connectors"Show help:
openwiki --helpIn chat, use /api-key to update the current provider API key and
/langsmith-key to update or clear LangSmith tracing credentials. Both commands
use masked prompts.
Authenticate a connector provider:
openwiki auth slack
openwiki auth gmail
openwiki auth x
openwiki auth notionStart an ngrok tunnel for Slack OAuth:
openwiki ngrok startThis starts ngrok with a random HTTPS forwarding URL. OpenWiki reads ngrok's
local inspection API, appends /callback, and saves
OPENWIKI_HTTPS_OAUTH_REDIRECT_URI automatically. Register the printed callback
URL in Slack. If you have a fixed ngrok domain, run
openwiki ngrok start https://<your-ngrok-domain>. X/Twitter and Gmail auth
ignore that HTTPS override and keep using the local loopback callback,
http://127.0.0.1:53682/callback.
Bare openwiki runs in code mode for the current repository. It creates initial repository documentation in openwiki/ when no wiki exists. Use openwiki personal for the local general-purpose wiki in ~/.openwiki/wiki/. By default, the CLI stays open after each run so you can send follow-up messages. Use -p or --print for a one-shot non-interactive run that prints the final assistant output.
Bare openwiki --init and openwiki --update default to code mode and operate on repository documentation. Use the personal positional mode or --mode personal to initialize or update the local personal brain wiki.
On each code run, openwiki maintains both an AGENTS.md and a CLAUDE.md at the repository root, adding prompting that instructs your coding agent to reference the wiki when searching for context. Each file is created if it does not already exist. If a file is present, OpenWiki only rewrites its own <!-- OPENWIKI:START -->…<!-- OPENWIKI:END --> block and leaves the rest of your content untouched (appending the block the first time). The scheduled GitHub Actions workflow includes these files, along with the workflow itself, in the documentation pull request.
Repository-specific wiki instructions are stored separately in
openwiki/INSTRUCTIONS.md. This file is a shared, user-authored brief for the
repository wiki: OpenWiki reads it for scope and priorities, but it is not
generated documentation and is not rewritten during normal init, update, or chat
runs unless you explicitly ask to change the brief.
On the first interactive run, OpenWiki will have you configure your inference provider, API key, and LLM. You will also be able to set a LangSmith API key to trace your OpenWiki runs to a LangSmith tracing project named "openwiki" (optional).
These configuration options and secrets will be saved to ~/.openwiki/.env on your local machine.
OpenWiki's first-run onboarding offers connector 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 local wiki under ~/.openwiki/wiki/ from those local files.
You can configure the same connector more than once. For example, add one Web
Search source for AI research and another for NBA news; OpenWiki stores them as
separate source instances such as web-search-1 and web-search-2. Run all
instances with openwiki ingest all, all instances for one connector with
openwiki ingest web-search, or one instance with
openwiki ingest web-search-2.
git-reporeads configured local repository paths and writes compact manifests.xuses the X API directly with OAuth user-context credentials for home timeline, user posts, mentions, bookmarks, and list posts.notiontargets the hosted Notion MCP server, so users should authenticate through Notion OAuth instead of pasting a Notion token into OpenWiki.googleuses the Gmail API directly with OAuth user credentials to fetch recent mail, with room to add Drive, Calendar, and other Google providers later.web-searchuses Tavily through LangChain and requiresTAVILY_API_KEY.hackernewsuses public Hacker News feed and search APIs, with no credentials required.
Connector secrets are referenced by env var name and stored in ~/.openwiki/.env; connector config files should never contain raw secret values.
openwiki auth <provider> runs a local browser OAuth flow, saves returned tokens into ~/.openwiki/.env, creates connector config when possible, and discovers MCP tools for MCP-backed providers. Slack and Gmail require app client credentials to already be set in that file; Notion uses dynamic client registration for hosted MCP; X uses OAuth 2.0 with PKCE. After openwiki auth gmail, the Google connector can ingest Gmail directly with no MCP transport setup.
openwiki auth configure <provider> and openwiki auth tools <provider> are advanced/retry commands for regenerating connector config or inspecting live MCP tools.
First-run onboarding also lets users choose a wiki template, customize its scope,
and save per-source ingestion notes and source schedules in
~/.openwiki/onboarding.json. The global personal wiki instructions are saved
in ~/.openwiki/INSTRUCTIONS.md. On macOS, source schedules are installed as
user LaunchAgents under ~/Library/LaunchAgents/ and write logs under
~/.openwiki/logs/.
See the OpenWiki operations docs for credential storage and provider setup notes.
OpenWiki supports OpenAI (with an API key or a ChatGPT login), OpenRouter, Gemini (AI Studio), Gemini Enterprise (Vertex AI), Nebius Token Factory, Fireworks, Baseten, NVIDIA NIM, an OpenAI-compatible provider, AWS Bedrock, and Anthropic out of the box. The onboarding default is OpenAI with gpt-5.6-terra, and each inference provider also includes pre-defined model options plus support for custom model IDs.
To route the Anthropic provider at an alternative, Anthropic-compatible endpoint
(for example a self-hosted or proxied gateway) instead of the default API, set
ANTHROPIC_BASE_URL alongside ANTHROPIC_API_KEY:
OPENWIKI_PROVIDER=anthropic
ANTHROPIC_API_KEY=your-key
ANTHROPIC_BASE_URL=https://your-gateway.example.com/anthropicThe openai-compatible provider targets any OpenAI-compatible chat-completions
endpoint via a required base URL. This can be used for OpenAI-compatible LLM
endpoints like those exposed by a LiteLLM gateway when it is used as a gateway —
letting you reach whatever upstream providers the gateway fronts through a single
OpenAI-shaped API. Set the model ID to whatever name the gateway exposes:
OPENWIKI_PROVIDER=openai-compatible
OPENAI_COMPATIBLE_API_KEY=your-gateway-key
OPENAI_COMPATIBLE_BASE_URL=https://your-gateway.example.com/v1
OPENWIKI_MODEL_ID=your-gateway-model-nameThe bedrock provider calls foundation models hosted on AWS Bedrock using IAM
credentials rather than a single vendor API key. It authenticates with an AWS
access key ID, a secret access key, and a region:
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-5Which model IDs are available depends on your AWS account and region (which foundation models you've enabled in the Bedrock console), so there is no preset model list — paste the Bedrock model ID directly, as shown above.
Some newer models only accept on-demand invocation through a cross-region
inference profile rather than their bare model ID — if you see ValidationException: Invocation of model ID ... with on-demand throughput isn't supported, prefix
the model ID with the profile's region code instead, for example
us.anthropic.claude-sonnet-5. Your IAM policy also needs to allow
bedrock:InvokeModel/InvokeModelWithResponseStream on both the
foundation-model and inference-profile resource types in that case.
The openai-chatgpt provider calls OpenAI's Codex backend using your ChatGPT
subscription instead of a metered API key. Model usage draws on your ChatGPT
Plus/Pro/Team plan's included Codex usage rather than per-token API billing. It
serves the same model list as the openai provider.
Instead of pasting an API key, run the setup wizard and complete a browser login:
OPENWIKI_PROVIDER=openai-chatgpt openwiki code --init
# or
OPENWIKI_PROVIDER=openai-chatgpt openwiki personal --initThe wizard opens https://auth.openai.com in your browser (and also prints the
URL for headless/SSH use, where you can open it on another machine — or paste the
redirect URL back into the terminal to finish without a callback). After you sign
in with your ChatGPT account, OpenWiki captures the OAuth callback, shows the
signed-in email and plan, and then continues to model and LangSmith selection
just like the other providers. It stores the resulting access token, refresh
token, expiry, account id, email, and plan in ~/.openwiki/.env
(OPENAI_CHATGPT_ACCESS_TOKEN, OPENAI_CHATGPT_REFRESH_TOKEN,
OPENAI_CHATGPT_EXPIRES_AT, OPENAI_CHATGPT_ACCOUNT_ID, OPENAI_CHATGPT_EMAIL,
OPENAI_CHATGPT_PLAN). These are managed for you — the access token is refreshed
automatically when it expires, so you normally never edit them by hand. Treat the
refresh token like a password.
The gemini provider runs Google's Gemini models through the AI Studio API with
a single API key:
OPENWIKI_PROVIDER=gemini
GEMINI_API_KEY=your-ai-studio-keyThe gemini-enterprise provider runs models from the Gemini Enterprise Model
Garden (formerly Vertex AI) — Google's own Gemini/Gemma models, Anthropic's
Claude, and partner/open-weight models (Llama, Mistral, DeepSeek, Qwen, …). It
routes each model ID to the right API surface automatically, so one credential
reaches all of them. It uses no API key — authentication happens with Google
Application Default Credentials (ADC), so any of the standard mechanisms work:
- a service account key file via
GOOGLE_APPLICATION_CREDENTIALS=/path/to/key.json, - user credentials from
gcloud auth application-default login, or - workload identity when running on Google Cloud (GKE, Cloud Run, GCE) or in CI.
OPENWIKI_PROVIDER=gemini-enterprise
GOOGLE_CLOUD_PROJECT=your-gcp-project
GOOGLE_CLOUD_LOCATION=global # optional, defaults to globalSet OPENWIKI_MODEL_ID to any Model Garden model. Gemini and Claude ship as
preset options; partner/open-weight models are reached by pasting their model ID
(for example publishers/meta/models/llama-3.3-70b-instruct-maas).
The credentials used need Vertex AI access (roles/aiplatform.user) in the
project, and the models you want must be enabled in the Model Garden. The
global endpoint serves Gemini and Claude and offers the best availability;
regional endpoints (for example europe-west1 or us-east5) can be set via
GOOGLE_CLOUD_LOCATION for data-residency requirements. Partner/open-weight
(MaaS) models are region-specific, so set GOOGLE_CLOUD_LOCATION explicitly when
using them.
Note that GOOGLE_CLOUD_PROJECT (and GOOGLE_APPLICATION_CREDENTIALS, if you
choose to store it there) is persisted to ~/.openwiki/.env and loaded into the
OpenWiki process environment at startup when not already set — values already
present in your shell always win.
For CI, authenticate before the update job runs — for example with
google-github-actions/auth
(workload identity federation) in GitHub Actions — and set
OPENWIKI_PROVIDER=gemini-enterprise and GOOGLE_CLOUD_PROJECT in the job
environment.
Base URLs (and all credentials) can be set in your environment or stored in ~/.openwiki/.env.
OpenWiki uses LangChain's built-in retry handling for transient provider errors.
To override the number of retries after the first provider request, set OPENWIKI_PROVIDER_RETRY_ATTEMPTS:
OPENWIKI_PROVIDER_RETRY_ATTEMPTS=3The value must be a positive integer. If the value is unset, OpenWiki defaults to 3 retries.
If there's an inference provider or model you'd like to see added, please open a PR!
OpenWiki collects anonymous, aggregate usage data so we can understand how the tool is used and improve it. Telemetry is on by default and easy to turn off.
What is collected, on a single openwiki_run event, keyed by a random
install ID stored locally in ~/.openwiki/install-id:
- Every run: the command (init / update) and the outcome (success / failure /
no-op), plus a coarse error category on failure (never the error message).
Interactive chat,
auth, andingestare not recorded. - At setup (on init only): which brain mode (code / personal), the model provider, and which connectors you configured (connector names only, never their contents).
What is never collected: file contents, repository data or names, credentials, prompts, model output, connector payloads, error messages, file paths, URLs, model IDs, run duration, your IP address, or any personal information. Geoip enrichment is disabled and your IP is never stored. Events are grouped by your random install ID so we can measure repeat usage, but that ID contains no personal data.
Scheduled/CI runs are collected as anonymous reliability data (tagged so
they can be told apart from human runs), under a shared CI identifier rather than
a per-machine install ID, and never counted as distinct installs. To disable in
CI, set OPENWIKI_TELEMETRY_DISABLED=1 in your workflow environment.
To see exactly what a run would send, add --telemetry-file=<path> to any run.
Set either environment variable:
export OPENWIKI_TELEMETRY_DISABLED=1
# or the cross-tool standard:
export DO_NOT_TRACK=1To disable permanently, add OPENWIKI_TELEMETRY_DISABLED=1 to ~/.openwiki/.env.
In CI, set it in the workflow environment (config files do not persist on
ephemeral runners).
Add --telemetry-file=<path> to any run to also write the exact payload to a
local JSON file.
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.