Show HN: Graft – local-first persistent memory for coding agents

Category: ai-ml

Tags: agent-memory, local-first, sqlite, embeddings, mcp, cli, retrieval, rag

Score: 7.3/10 (Innovation: 6, Technical: 8, Documentation: 8, Utility: 7)

Graft is a local-first persistent memory layer for AI coding agents, implemented in C11 with SQLite (FTS5 + sqlite-vec) and local BGE-M3 embeddings via llama.cpp. It offers three retrieval modes (query/retrieve/explore) with confidence gating, hybrid RRF retrieval, and graph exploration, plus integrations for Claude Code, Codex, Gemini CLI, ChatGPT, and MCP. It's an interesting take on agent memory that avoids SaaS and external embedding APIs while filling a real gap in cross-session coding agent workflows.

Target audience: AI/ML engineers and backend developers building coding agents or RAG-adjacent developer tooling

Repository: https://github.com/AEndrix03/Graft · C · Apache-2.0 · 17 stars

View on Hacker News