NoStale extracts knowledge from your codebase, indexes it, and generates PRD/TDD documents that stay in sync with every commit, no stale docs, no manual regeneration. Browse it from a web UI; ask it questions from the coding agent you already use, over MCP.
A pipeline triggered by git extracts your codebase (AST-based, multi-language), indexes it for retrieval, and regenerates only the PRD/TDD documents affected by what changed, each version reviewable before it replaces the last. A web UI handles browsing (text and graph view) and configuration, repos, LLM providers, document templates. It deliberately has no chat feature: AI Q&A goes through an MCP server instead, so you ask questions from Claude Code, Cursor, or any MCP-compatible agent, not a second chat window.
npm install -g @nostale/cli
nostale setupNo prior clone needed — the wizard walks you through personal mode (no Docker) or server mode (full docker-compose stack, images pulled from GHCR), writes your config, runs migrations, and in personal mode provisions your first account automatically.
nostale status # what's running
nostale stop # stop everythingUpgrading later is the same one-liner: npm install -g @nostale/cli@latest && nostale stop && nostale start — your existing ~/.nostale data and config are untouched; rag-core's migrations re-run automatically on wiki-api's next boot, and release notes call out when a new required env var means you should re-run nostale setup instead.
git clone https://github.com/fetot/nostale.git && cd nostale
npm install
npm run build --workspace=@nostale/cli
node apps/cli/dist/index.js setup- Incremental sync, only re-extracts and re-generates what actually changed, not the whole repo on every push.
- PRD/TDD generation, versioned, reviewable documents, customizable templates per project or globally.
- MCP integration,
search_wikiandask_wikireachable from any MCP client, no separate chat UI to maintain. - Two deployment modes, personal (no Docker, one command) or server (Docker Compose, for a team sharing one wiki).
- Bring your own LLM, Anthropic or any OpenAI-compatible endpoint, including self-hosted ones.
extract → index → generate (wiki-worker, the only service that calls an LLM)
│
libSQL (file, or sqld)
│
┌────────┴────────┐
wiki-api mcp-server (auth + REST) (the only path for AI Q&A)
│
wiki-web (browse + configure, no chat)
A general-audience write-up lives on nostale-site, see below.
Full docs (getting started, installation, MCP integration, configuration reference, FAQ) live on nostale-site, a separate repo kept lean so this one stays focused on the product. It isn't deployed yet, this section will link to it once it is.
See CONTRIBUTING.md for dev setup and how work here is scoped, and CODE_OF_CONDUCT.md for community guidelines.