The AI assistant that never forgets
Most AI assistants reset between sessions. Nuggets solves this with a three-layer memory system that stays fast, stays local, and gets smarter over time.FHRR Fact Memory
Algebraic recall in under a millisecond — no embeddings, no vector database required.
Living Note Graph
Zettelkasten-style notes with titles, tags, links, and FHRR similarity search.
Autonomous Rewriting
Daily cleanup pass that merges duplicates, improves tags, and archives stale notes.
Get up and running
Run setup
Full setup guide
Step-by-step instructions including prerequisites, environment variables, and first-message verification.
Everything in one place
Messaging Gateway
Connect Telegram, WhatsApp, and Discord to a single AI backend.
Backend Modes
Choose between Pi, Codex CLI, or a local OpenAI-compatible model.
Skills
Extend the assistant with scoped, triggerable skill files.
API Reference
Full TypeScript API for the Nugget class, NuggetShelf, graph notes, and rewrite engine.
How memory works
Nuggets separates memory into three cooperating layers:| Layer | Best for | Storage |
|---|---|---|
| Fact memory | Short key-value facts recalled by query | .nugget.json files |
| Note graph | Rich notes with tags, links, and context | ~/.nuggets/graph/graph.json |
| MEMORY.md | Permanent, high-value knowledge | Promoted from the graph |
All memory is local. No cloud database, no embeddings API — just fast algebraic math and JSON files on your machine.
Architecture overview
Core concepts
Understand how the three memory layers work together.
Configuration reference
Every environment variable, with defaults and examples.