Built for how engineers work
Every capability purpose-built for performance — from the storage engine to the graph traversal.
Up and running in under a minute
No accounts. No configuration. Just install and go.
Install
One command to install.
Initialize & Index
Point it at your project.
Search & Build
Start querying your knowledge.
Connected knowledge
Documents, tasks, and code linked by typed relationships — what blocks what, what implements what, what broke what. Your agent navigates the full picture instead of guessing.
- Wikilinks and automatic backlinks
- Typed edges: references, implements, blocks, resolves, parent_of
- Code symbol linking across 17 languages
- Commit references link KB documents to git history
Understands your code
Agents get 3 exact call sites instead of 200 grep matches. Index your codebase across 17 languages. Call graphs, impact analysis, and dead code detection — before anything gets changed.
- AST-based symbol extraction via tree-sitter
- Full call graph: callers, callees, transitive impact
- Dead code detection — find symbols with zero callers
- Auto-reindexing via file watcher (500ms debounce)
Every change is tracked
Every edit is a versioned commit with full diffs. Atomic multi-document operations, crash-safe writes, and a complete audit trail. Nothing is silently lost or overwritten.
- Atomic multi-document commits
- BLAKE3 cryptographic integrity verification
- Full commit log with structured diffs
- Audit trail for compliance
Measurably instant
Every query, every traversal, every lookup returns before you notice it left. Built in Rust with SQLite and FTS5 — cold start included.
- Full-text search with BM25 ranking via FTS5
- Sub-25ms search across thousands of documents
- Zero startup latency — no daemon required for core operations
Sync only what you need
Checkout individual documents to an editing workspace. Push and pull per-document — not the whole database. Stubs represent documents you haven't pulled yet, preserving graph integrity without the weight.
- Per-document checkout to editable workspace
- Stubs for unsynced documents — graph stays connected
- Offline-first — works without any network connection
- Efficient bandwidth — only changed documents transfer
- Isolated workspaces per git worktree — many agents, per-repo; zero conflicts
42 MCP tools & agent plugins
One command configures your agent with rules, skills, slash commands, and MCP tools. Works with Claude Code, Codex, and any agent that speaks the MCP protocol.
- 3 rules, 12 slash commands, 4 skills, 2 config files
- 42 MCP tools for search, graph, code intel, and more
- Persistent context across agent sessions
- Agent-agnostic: Claude, Codex, Cursor, any MCP client
80+ commands that feel like git
Every change is a commit — full auditability for every agent action across the distributed knowledge base. No vendor lock-in, no GUI dependency. Pipe, script, alias. Works in any terminal, any CI, any automation.
- Document management: create, show, list, edit, delete
- Code intelligence: index, callers, callees, impact, dead-code
- Task management: board, set, link
- Sync: push, pull, checkout, commit, status, diff, log
17 languages and counting
Each language pack provides AST-based symbol extraction, call graph analysis, and dead code detection via tree-sitter grammars.
Spec-driven development at organizational scale
GitKB enables a fundamentally new engineering workflow. Engineers iterate on specs and tasks — the thought work — then dispatch them for concurrent implementation by agents. The result is massively parallelized development where humans design and agents build.
Design
Engineers write specs and break them into tasks with clear acceptance criteria. GitKB connects every spec, task, incident, runbook, or custom typed document to the existing knowledge graph — and to the architecture, code, and decisions that aid in informing them.
Dispatch or dive in
Refine a task to full confidence and dispatch it to an agent in the background — or dive in and implement a complex task in close conversation with your agent. Either way, the full context loads automatically: the spec, the related code, the call graph, the conventions.
Land
Engineers review, iterate, and land the work. Every completed task feeds back into the knowledge graph — expanding context for the next round. The cycle accelerates.
See it in action — one prompt, and your agent uses code intelligence to build a complete implementation plan:
Code intelligence indexing means the agent queries the AST and call graph directly — not grepping through files. Precise results at a fraction of the token cost, both for discovery and for dispatching well-scoped tasks.
This is how engineering organizations eliminate repetitive work and amplify thought work. Engineers become architects and reviewers. Agents become the builders. GitKB is the connective tissue that makes it work.
"Do we have a spec for that?"
Specs, tasks, incidents, code — all searchable, all linked. Your agent finds what exists before creating something new.
Get started
Local and free forever. No account required. Install and start building in under a minute.
Evaluating for your team? Join the Alpha