contextlake 2.23.0

A local context layer for AI tools: mirror your repositories, index them into a knowledge graph, and serve it over MCP so agents answer from real source instead of guessing.

contextlake

All your real context, in one local lake.

A local context layer for your AI tools: mirror your repositories, index them
into a knowledge graph, and serve it over MCP, so agents answer from real source instead of guessing.

---

Why contextlake

Your AI assistant is only as good as what it can actually see. Point it at one file and it's sharp; ask it about the system, which service calls this API, who depends on that package, where a symbol is really defined across dozens of repos, and it starts guessing.

contextlake gives your tools the real source to read. It mirrors your repositories to your machine, indexes them into a queryable knowledge graph, and serves that graph to your editor over MCP. Everything runs locally and offline, no code leaves your machine, and it carries no credentials of its own.

How it works

contextlake is three layers you adopt one at a time. The mirror is useful on its own, and each layer above it is optional.

1. Mirror: clone every repo you can reach in a GitLab group into a faithful copy of its namespace tree, each on its most active branch, kept fresh with one command. (The source is GitLab today; the design is source-agnostic.)
2. Knowledge layer (optional): parse the mirror into a code + dependency graph, add semantic search, a council-verified wiki (each page reviewed and scored before publishing, low-confidence pages dropped), and connectors to Atlassian / Figma / GitLab.
3. Serve: expose it all over MCP and an offline interactive graph visualizer, so agents can answer "where is X defined?" or "who calls Y?" instead of grepping.

Each layer has its own guide: the mirror in Usage & config, the knowledge layer and serving in Knowledge layer, and the whole flow start to finish in QUICKSTART.

Install

pip install "contextlake[kb]"       # the full tool: mirror + graph, search, wiki, MCP server
pip install contextlake             # mirror-only core (no pip dependencies at all)

Everything in the quickstart below needs the [kb] extra (Python 3.10+); the plain install is just the mirroring CLI and runs on Python 3.9+.

Prefer an isolated, zero-setup install? uv fetches the right Python and an isolated environment for you:

uv tool install "contextlake[kb]"            # install the CLI on your PATH
uvx --from "contextlake[kb]" contextlake --help   # …or run it once, without installing
# pipx install "contextlake[kb]"             # pipx works too

Install extras (the mirror needs none — add these for the knowledge layer)

Extra	Adds	When you need it
[kb]	The knowledge layer: parse → graph → wiki → MCP server	Anything beyond mirroring
[kb-full]	[kb] + the built-in CPU embedder + sqlite-vec ANN	One-step local semantic search, no Ollama or API key
[kb-vec]	The sqlite-vec ANN backend	Faster vector search than the pure-Python fallback
[kb-local]	The built-in CPU embedder (model2vec, ~30 MB)	Semantic search with no Ollama or API key
[kb-fastembed]	A higher-quality ONNX embedder (~90 MB)	Better semantic ranking
[llm-local]	A built-in CPU model for the wiki (llama-cpp)	wiki --llm builtin with no Ollama or API key

Docker (turnkey / air-gapped: models baked in)

The published image bundles the knowledge layer plus the built-in CPU models (embedder + a small wiki LLM), so it runs with no Ollama, no API key, and no model download at runtime. The PyPI wheel stays the primary install; reach for the image on locked-down or offline machines.

docker run -v "$PWD:/work" ghcr.io/sayak-sarkar/contextlake doctor
docker run -v "$PWD:/work" ghcr.io/sayak-sarkar/contextlake index

From source (for contributors)

git clone https://github.com/sayak-sarkar/contextlake && cd contextlake
pip install -e ".[kb]"

Prerequisites: git, and, only for GitLab mirroring, either a GITLAB_TOKEN env var (a PAT with read_api + read_repository — no other tool needed) or an authenticated glab (glab auth login). The knowledge layer needs neither. Once installed, contextlake, python -m contextlake, and python3 contextlake.py are equivalent.

Quickstart: one repo, no setup

You don't need GitLab or any config to try contextlake on a repo you already have. No install? Run it once with uvx: prefix any command below with uvx --from "contextlake[kb]" (e.g. uvx --from "contextlake[kb]" contextlake index --source .).

contextlake index                     # parse the current repo into a local knowledge graph
contextlake graph --overview --open   # open the interactive graph in your browser
contextlake serve                     # …or serve it to your AI IDE over MCP

Wire it into your editor in one line, no config file needed (it uses the local ~/.contextlake/kb store you just built):

claude mcp add contextlake-kb -- contextlake serve      # Claude Code
# zero-install variant: claude mcp add contextlake-kb -- uvx --from "contextlake[kb]" contextlake serve

contextlake graph, a whole codebase as one offline, navigable graph.

Everything lands in a local store (~/.contextlake/kb), nothing leaves your machine. Index any path with --source PATH, or every git repo under a directory with --workspace DIR.

Want the full path, mirror a GitLab fleet → graph → wired editor in a few minutes? QUICKSTART.md walks the whole flow.

Fleet mode: mirror a GitLab group

Where contextlake goes beyond single-repo tools is mirroring and cross-referencing a whole GitLab fleet. Copy the example config and set your group + workspace:

cp .contextlake.ini.example ~/.contextlake.ini

[contextlake]
work_dir = ~/work
gitlab_group = your-gitlab-group

contextlake status      # see where you stand (read-only)
contextlake sync        # fetch → clone → update → branches → verify → audit

It carries no credentials of its own (auth rides on your existing glab login), so .contextlake.ini holds only non-secret settings and is gitignored by default. It runs across hundreds of repos concurrently, with an adaptive worker pool, retries with backoff, and never stomps on the feature branch you're in the middle of.

Behind a slow / TLS-inspecting corporate proxy (e.g. Zscaler) where glab's API calls time out? Set GITLAB_TOKEN (a read_api token) and contextlake enumerates projects via its own HTTP client, which tolerates the slow DNS where glab's short dial timeout fails.

Commands at a glance

Run any command as contextlake <command>; each has scoped help via contextlake <command> --help. Per-command docs live with their layer: the mirror commands in usage.md; the knowledge-layer commands (index, embed, connect, wiki, query, owners, impact, graph, …) in knowledge-layer.md, and serve/steer in serve.md.

Command	What it does
status	Show the workspace sync state vs GitLab (read-only)
sync	The full pipeline: fetch → clone → update → branches → verify → audit
fetch · clone · update	The sync steps, individually
branches	Switch each repo to its most active branch
verify · audit	Check the mirror vs GitLab; report repo health, age & drift (JSON + CSV)
bootstrap	Turnkey: sync + index + connect + embed + wiki + steer
index	Build the code/dependency graph (--workspace, incremental, --watch)
connect	Link repos to Atlassian / Figma / GitLab items (--watch to keep refreshing)
embed	Build semantic-search vectors (zero-config built-in CPU model, Ollama, or an API; incremental, --watch)
ingest	Aggregate external docs into the graph + semantic store (built-in files/web/api/mcp sources, or plugins)
wiki [<repo>…]	LLM-synthesized, council-verified wiki pages (all repos, or just the named ones); --llm builtin|ollama|openai enables the LLM tier inline
query	Search the index (--kind, --repo, --as-of <commit>)
owners (alias who-knows)	Likely owners / SMEs for a repo (or --path), ranked from git history
impact (alias blast-radius)	Change-impact / blast radius: what depends on a symbol (--hops, --repo to disambiguate)
graph	Visualize the graph, offline interactive HTML / DOT / Mermaid / JSON
dashboard	Local knowledge-system dashboard UI (--serve; --sample for the bundled demo fleet; --site DIR for a static offline export)
serve	Expose the graph over MCP (--transport stdio/http)
steer	Write editor steering, AGENTS.md, .mcp.json, .windsurfrules, skills
lint · doctor · eval	Graph health · environment check · retrieval-quality scoring

Global options apply to any command: --dry-run (preview without changing anything), -v/-q (verbosity), --log-file PATH, --config PATH, --version. Output is colorized on a TTY and plain when piped; set NO_COLOR to force-disable.

Knowledge layer

Beyond mirroring, the optional contextlake.kb layer turns your repos into a knowledge graph and serves it to AI tools over MCP. It can link repos to their Atlassian / Figma / GitLab items, add semantic search, write a curated wiki, visualize the graph (offline interactive HTML, fleet overview, a symbol's neighbourhood, or a single repo), and generate per-tool steering files + a skills library. Most of it needs no model; the rest works with a local Ollama or any OpenAI-compatible endpoint.

One command sets it all up (configs are read from their default locations):

contextlake bootstrap

Full guide: docs/knowledge-layer.md.

The dashboard

contextlake dashboard --serve opens a local, offline-first window into everything the knowledge layer builds — a fleet overview, per-repo anatomy, the cross-repo architecture graph, change-impact (blast radius), health, and search. Try it with zero setup via contextlake dashboard --serve --sample.

The dashboard — a guided tour — step by step, with screenshots.

Documentation

• QUICKSTART.md, install → bootstrap → wire your editor, in minutes
• docs/dashboard.md, the dashboard, a guided tour with screenshots
• docs/usage.md, every command, configuration, branch safety, scheduling
• docs/knowledge-layer.md, the graph, connectors, search, wiki
• docs/serve.md, serve the graph over MCP + wire your editor
• docs/internals.md, architecture & internals
• docs/releasing.md, maintainer runbook: versioning, tagging, publishing
• CHANGELOG.md · ROADMAP.md · CONTRIBUTING.md · BRANDING.md

License

MIT, see LICENSE. Pebble the otter is the project mascot; deep context, clear answers.