EN
Memo Documentation
Welcome to the Memo documentation. Memo is a local-first, privacy-focused LLM chat application with a semantic RAG memory engine, agent tools, and multi-model orchestration. It runs entirely on your machine — no telemetry, no cloud dependency, no training on your data.
Current version: v4.5.0
What is Memo?
Memo is your AI second brain. It runs LLMs locally via a bundled llama.cpp runtime, remembers everything you discuss in a SQLite + sqlite-vec + FTS5 vector store, and can use external cloud AI providers when you need more power. The choice is yours — per model, per conversation.
Core Capabilities
| Capability | Description |
|---|---|
| RAG Memory | Semantic vector search with hybrid retrieval (vector + FTS5) via Reciprocal Rank Fusion |
| Agent Mode | 27 built-in tools — file operations, shell commands, web search, routines, WhatsApp messaging |
| Code Mode: Plan / Auto / Build | Three gears for agentic coding, cycled with Ctrl+Tab — Plan investigates and writes a saved plan, Auto confirms edits, Build runs without waiting |
| Self-Driving Task Loop | Hand Memo a Task.md checklist and walk away — planner/executor mode, up to 3 parallel sub-agents, escalating retry, notifies on every outcome |
| Desktop Mascot | A small always-on-top character, its own window, reflecting Memo's live activity across chat, WhatsApp, Telegram, and the task loop |
| Live Mode v2 | Native audio-to-audio voice via Google Live or OpenAI Realtime, with barge-in and mid-conversation memory recall |
| Orchestra Mode | Multi-model parallel workflows with a chief model coordinating 8 specialist roles |
| 16 Providers | OpenAI, Claude, Gemini, Grok, Groq, OpenRouter, Ollama, OpenCode Zen, OpenCode Go, Kilo Code, gemini-sub — with smart fallback chains and a free/paid model browser |
| WhatsApp & Telegram | QR-pair WhatsApp or connect a Telegram bot — message yourself and Memo replies as a full assistant, agent tools included |
| Routines | Natural-language scheduled automations — now creatable, listable, and cancellable straight from a WhatsApp or Telegram conversation |
| Cloud Sync | E2E-encrypted Google Drive backup (AES-256-GCM, PBKDF2) |
| Mobile Companion | Thin Flutter app for Android/iOS — LAN, ngrok, or Tailscale tunneling; the self-hosted web UI is fully mobile-responsive too |
| Proactive Learning | Intent extraction from conversations, automatic calendar events, mood engine |
| Self-Hosting | Server-only install for a Pi/home server/VPS, Docker/CasaOS, 4-mode auth, multi-account with 7 independent per-account permissions, and a full memo config/remote/service/provider/agent/model CLI toolkit over SSH — no display required |
Quick Links
Getting Started
- Quick Start Guide — Two terminals, up in 60 seconds
- Installation — One-click installers for every platform
- System Requirements — What your machine needs
- First Run Wizard — Setup walkthrough
- Build from Source — Compile from the repo
- Self-Hosting — Server-only install, Docker/CasaOS, auth modes, CLI management
Core Concepts
- What is Memo? — Philosophy and full feature overview
- How RAG Memory Works — Embedding, retrieval, and persistence
- System Architecture — Two-process design and data flow
- Backend Architecture — Go backend deep dive
- Frontend Design — Flutter UI and state management
- Data Layer — SQLite, file layout, and persistence
Advanced Features
- Agent Mode — Tool-calling sandbox
- Orchestra Mode — Multi-model workflows
- External Providers — Cloud AI configuration
- Proactive Learning — Intent extraction and calendar
- Cloud Sync — Encrypted Google Drive backup
- WhatsApp Integration — Bridge setup and usage
- Telegram Integration — Bot setup, owner lock, self-chat assistant
- Mobile App — Remote companion
Development
- Development Guide — Architecture, conventions, tooling
- API Reference — REST + SSE endpoints
- CLI Reference —
memo config/remote/service/provider/agent/model - CGO Flags — SQLite compilation requirements
- Build & Packaging — Releases and distribution
- Contributing — How to help
Reference
- Configuration — config.yaml reference
- Advanced Settings — All tuning knobs
- Technical Deep Dive — Internals explained
- UI Features — Interface tour
- Security — Encryption, permissions, and threat model
Resources
- Roadmap — What's coming next
- Troubleshooting — Common problems and solutions
- Release Notes — Version history
Memo is GNU AGPL v3 licensed. View on GitHub