ContextForge

Work in Progress

AI coding assistants forget everything between sessions. ContextForge fixes this with an MCP server that provides persistent semantic memory — hybrid search combining keyword matching and vector embeddings, automatic codebase analysis, and intelligent context recovery.

Hybrid search — FTS5 keywords + vector embeddings find context by meaning, not just words
Code intelligence — tree-sitter parses structure, exports, and dependencies across 50+ languages
Git-aware — extracts architectural decisions from conventional commits automatically
Zero config — single binary, brew install, works offline

Built in Rust with libSQL for storage and local embeddings via candle — no external APIs required.

Rust
MCP Protocol
libSQL
Vector Search
tree-sitter

View Source Code

Key Features

Hybrid semantic + keyword search (FTS5 + vector embeddings)
Automatic codebase analysis with tree-sitter (50+ languages)
Git commit parsing for architectural decision extraction
MCP native — plugs into Claude Code, Cursor, Copilot
Single Rust binary, ~4-8MB, zero external dependencies