--- url: https://fewknowme.github.io/ctxpilot access_date: 2026-07-23T22:58:28.000Z current_date: 2026-07-23T22:58:28.847Z --- # Your AI already knows your codebase. ctxpilot builds a living summary of your project and injects it into every AI tool you use. Run it with Anthropic, OpenAI, or free local Ollama models. $ `npm install -g @ctxpilot/ctxpilot` MIT License TypeScript MCP Protocol New: local Ollama Anthropic + OpenAI Open Source --- New in v0.1.2 ## Free local context generation with Ollama Choose local mode during `ctx init`, select a pulled Ollama model, and keep project context on your machine while ctxpilot builds and updates your LCD. **Guided setup** Detects Ollama, lists local models, and saves the selected model. **Manual config** Use `CK_PROVIDER=local`, `CK_MODEL`, and optional `CK_LOCAL_URL`. **Better resilience** Repairs near-JSON model output and archives parse failures without stopping updates. Release notes npm package --- What it looks like \~ ctx init How would you like to run ctxpilot? 1) Free (local) - Ollama, data stays on your machine 2) BYO API key - Anthropic or OpenAI Selected model: gemma3:4b Scanning project... reading git log, package.json, source files LCD generated. 847 tokens. .ctxpilot/context.md written. \~ ctx setup Configured: Claude Code, Codex CLI, Cursor, Windsurf CLAUDE.md written. Codex skill created. Cursor + Windsurf rules written. \~ ctx watch Daemon started. Watching .git/refs/heads/ for commits... Auto-update will run 30s after each commit. Fresh Codex session — zero context pasted Asked:**"What is this project and what am I working on?"** Codex responded: "Minimum per-item quantity is hardcoded as **6** in several places, and the store silently coerces values with**Math.max(quantity, 6)** in **cart.ts#L40** and **cart.ts#L70**." Exact file. Exact line numbers. Private codebase. Zero explanation given. --- ## Three commands to set up Step 1 `ctx init` Scans your codebase with AI. Builds a Living Context Document from your files, git history, and answers to 3 quick questions. Step 2 `ctx setup` Hooks into every AI tool you have installed. Writes MCP configs, CLAUDE.md, Cursor rules, and Codex skills automatically. Step 3 `ctx watch` Starts a background daemon. Detects git commits. Auto-updates the LCD. You never run ctx update manually again. --- ## Works with your tools ctx setup auto-detects installed clients and writes configs in one command. Claude Code MCP + CLAUDE.md Cursor MCP + .cursor/rules/ Codex CLI MCP + agent skill Windsurf MCP + .windsurf/rules/ --- ## Commands | ctx init | Scan codebase and build the first LCD | | ----------------- | ---------------------------------------------------------- | | ctx setup | Write MCP configs and rule files for all installed clients | | ctx watch | Start background daemon, auto-update on git commits | | ctx update | Run one incremental update now | | ctx build | Rebuild LCD from scratch, archive previous | | ctx show | Print current LCD with token count | | ctx inject | Output LCD as XML/markdown/plaintext for manual paste | | ctx signal "text" | Add a manual note for next update | ---