Personal AI Sovereignty
A local-first AI assistant with 20 integrated tools, long-term memory, voice, vision, health tracking, and messaging channels โ all running on your machine. Your models, your data, your rules.
Governments are investing billions to keep AI infrastructure within their borders. Thoth applies the same principle to the individual โ your compute, your data, your choice of model, accountable to no one but you.
No cloud dependencies. No subscriptions. No data leaving your machine.
Every LLM call, every voice transcription, every memory lookup runs on your hardware. No data is sent to any cloud AI provider โ ever. External APIs are opt-in and individually toggleable.
Core PrincipleThe agent reasons about your question, decides which of 20 tools to use, chains multiple lookups, and assembles a complete answer โ all autonomously. Smart context trimming keeps multi-tool workflows within window limits. Live token counter tracks usage in real time.
Thoth remembers what matters โ names, birthdays, preferences, projects โ across every conversation. FAISS semantic search with automatic recall injects relevant memories before every response. Background extraction periodically scans past chats for personal facts you mentioned in passing.
Track medications, symptoms, exercise, periods, mood, sleep โ through natural conversation. "I took my Lexapro" โ logged. 7 built-in analyses: adherence, streaks, trends, co-occurrence. Results auto-export to CSV and chain to interactive Plotly charts.
New in v3.0Connect a Telegram bot or email channel so the agent meets you where you are. Every message is processed by the full ReAct agent with all tools available. Each chat gets its own conversation thread.
New in v3.0Local speech-to-text via faster-whisper (CPU, int8 quantized) and neural TTS via Piper with 8 voice options. Mic auto-mutes during playback. Fully hands-free, fully offline โ no cloud APIs touch your voice data.
Capture images from your webcam or take screenshots and ask questions about them. Local vision model analysis โ multi-camera support, inline image display in chat.
Build reusable multi-step prompt sequences that chain research โ summarisation โ action. Run manually or schedule daily/weekly. Template variables, background execution, desktop notifications, and 4 built-in templates to start.
One-click installer, native OS window via pywebview, system tray with status indicator, branded splash screen, and a first-launch setup wizard that walks you through model selection. No Docker, no YAML, no terminal needed.
Switch freely between Qwen, Llama, Gemma, Mistral, DeepSeek, Phi, Nemotron, and more โ only models verified for tool calling are included. Download buttons with live progress. Configurable context windows up to 256K tokens with automatic model-max capping.
Search emails, read threads, draft and send messages. View, create, update, and manage calendar events โ all through natural language. Tiered permissions let you control read, write, and destructive access independently.
Sandboxed file operations with PDF, CSV, Excel, JSON support. Structured files return schema + stats + preview via pandas. Interactive Plotly charts rendered inline โ bar, line, scatter, pie, heatmap, and more. Destructive actions require confirmation.
The agent autonomously decides which tools to use based on your question
Cloud AI rents you access to someone else's system. Thoth gives you ownership.
| Cloud AI (ChatGPT, Claude, Gemini) | ๐ Thoth | |
|---|---|---|
| Your data | Sent to cloud servers, used for training | Stays on your machine โ always |
| Conversations | Owned by the provider, can be leaked | Local SQLite, fully yours |
| Cost | $20+/month per subscription | Free forever โ your hardware |
| Memory | Limited, opaque, provider-controlled | You control what's remembered |
| Tools | Sandboxed plugins, limited access | Gmail, Calendar, filesystem, webcam |
| Voice | Cloud-processed speech | Local Whisper + Piper TTS |
| Health tracking | None | Conversational tracker + analysis |
| Availability | Requires internet, subject to outages | Works offline, no throttling |
Run the one-click Windows installer โ it sets up Ollama, Python, and all dependencies automatically. Or clone the repo and run from source in 5 commands.
On first launch, a setup wizard walks you through selecting and downloading a brain model and vision model. The default (qwen3:14b) is recommended for the best balance of speed and capability.
Ask anything in natural language โ the agent figures out which tools to use. Search the web, check your email, analyze your screen, track a medication, set a timer. Just ask.
Thoth doesn't just promise privacy โ it enforces it structurally. All LLM inference, voice processing, memory storage, and conversation history runs entirely on your machine. External network calls only happen when you explicitly use an online tool โ and each one can be individually disabled.
Free, open source, and yours to keep. No accounts. No subscriptions. No data harvesting.