buildingjoshbetter/TrueMemory

The memory your AI should have had from the start. Automatic capture, automatic recall, 100% local. One SQLite file, zero cloud. Works with Claude Code, Claude CLI, Cursor, Codex CLI, Gemini CLI.

What it solves

TrueMemory provides a long-term, local-first memory system for AI agents, preventing the "amnesia" that occurs when a new session starts. It filters out noise from thousands of messages to identify key information, updates stale facts and resolves contradictions as the user's preferences or decisions change over time.

How it works

It uses a retrieval-centered architecture with a 6-layer pipeline and a reranker to find relevant memories. The system operates in three tiers (Edge, Base, and Pro) depending on theon available hardware and desired accuracy. It stores all data locally in a single SQLite file and can be integrated as a Python library or via lifecycle hooks for tools like Claude Code, Cursor, and Gemini CLI to automatically capture and inject memories into sessions.

Who it’s for

Developers building AI agents or users of AI coding tools (like Cursor or Claude CLI) who want their AI to remember preferences, technical stacks, and past decisions across multiple sessions without manual tagging or prompt engineering.

Highlights

  • Local-first privacy: All memories are stored in a local SQLite database, ensuring data never leaves the device.
  • Automatic capture: Automatically extracts and injects relevant memories from conversations without manual input.
  • Detailed Tiers: Offers Edge (CPU only), Base (fully offline), and Pro (AI-powered query expansion) options.
  • High Performance: Achieves SOTA results on the BEAM-1M benchmark for long-term recall.
  • Agentic Search: Includes a search_deep method for multi-round, higher-accuracy retrieval.