buildingjoshbetter/TrueMemory

curl -LsSf https://raw.githubusercontent.com/buildingjoshbetter/TrueMemory/main/install.sh | sh (Claude Code / Claude CLI / Cursor / Codex CLI / Gemini CLI; downloads ~1.5GB of AI models into an isolated env)
Windows PowerShell: irm https://raw.githubusercontent.com/buildingjoshbetter/TrueMemory/main/install.ps1 | iex
Python library: pip install truememory

Usage

• •After installing, quit and reopen your AI tool, then type 'Set up TrueMemory' and pick a tier
• •Lifecycle hooks capture conversations automatically — no manual work needed
• •Python API: from truememory import Memory; m = Memory(); m.add("Prefers dark mode", user_id="alex"); m.search("preferences", user_id="alex")
• •Switch tiers anytime by saying 'switch to Pro' or 'switch to Base'

Key features

• ✓Automatic capture and injection of memories across sessions; directives (standing instructions) auto-load every session
• ✓100% local: one SQLite file at ~/.truememory/memories.db; Edge and Base tiers make zero external calls
• ✓Three tiers in one install: Edge (CPU-only, 8MB embedding model), Base (fully offline, 600MB model), Pro (adds HyDE query expansion, needs LLM API key)
• ✓6-layer retrieval pipeline + reranker; search_deep multi-round agentic search
• ✓Reported 93.0% LoCoMo / 92.0% LongMemEval / 76.6% (SOTA) BEAM-1M with reproducible benchmark scripts
• ✓Works via lifecycle hooks with Claude Code, Claude CLI, Cursor, Codex CLI, Gemini CLI, Claude Desktop

Best for

Developers who want zero-config, private, automatic long-term memory for their coding agent without managing a server or cloud account

Caveats

• ⚠Pro tier requires an LLM API key for HyDE query expansion (Pro sends only your search query text to the LLM, never memories)
• ⚠Installer downloads ~1.5GB of local AI models; first run takes 3-5 minutes
• ⚠Anonymous usage telemetry is on by default (never memory content); opt out with TRUEMEMORY_TELEMETRY=off
• ⚠AGPL-3.0: free for personal and research use; commercial use requires a separate license