TaskWing — Give Your AI Tools a Brain
Your AI forgets everything every session. Context compression tools save tokens. TaskWing saves knowledge — decisions, patterns, and architecture that compound across every session.
Install: brew install josephgoksu/tap/taskwing or curl -fsSL https://taskwing.app/install.sh | sh
The Problem: AI Has No Memory
~15 min/session re-explaining your stack. 62% of AI suggestions ignore your patterns. Context dies when the session ends. The real cost isn't tokens — it's knowledge loss.
How It Works: Extract. Connect. Compound.
Extract — One command scans your codebase and extracts decisions, patterns, and constraints into a local SQLite database. No cloud. No signup.
Connect — Exposes your architecture to Claude, Cursor, Copilot, and Gemini via the Model Context Protocol (MCP). Your AI queries it mid-conversation.
Compound — Every decision, every pattern, every constraint — persisted permanently. Session 50 is smarter than session 1.
Benchmark Results: 3.6 → 8.0
Same prompts, same model, same codebase. The only difference: TaskWing gave the AI your architecture. +122% improvement on architectural coding tasks.
Why Not Just Use CLAUDE.md?
CLAUDE.md is a great start. TaskWing is what happens when your project intelligence becomes permanent, searchable, and compounding. It auto-extracts from code, grows over time, and works across AI tools via MCP.
Open Source. Zero Lock-in.
MIT Licensed. Works offline. No account required. Local SQLite storage. Your architecture stays local.
View on GitHub · Built by Joseph Goksu