Local-First AI Assistant
Thoth is a desktop AI assistant with memory, tools, workflows, design creation, messaging, plugins, MCP tools, and provider-aware model routing. Run it fully local with Ollama, add frontier APIs, or sign in to ChatGPT / Codex when you want subscription-backed models. Your durable data stays on your machine.
Thoth is built for personal AI sovereignty: your compute, your data, your model routes, and your automation under your control.
Create decks, documents, landing pages, app mockups, and storyboards with live preview, surgical AI edits, media generation, chart insertion, export, and shareable interactive links.
Five modesMemories become connected entities and typed relationships, with semantic recall, graph expansion, visual exploration, document extraction, and Obsidian-compatible wiki export.
Durable memoryThe agent can browse in Chromium, use the shell, manage files, search the web, read documents, send email, manage calendar events, inspect status, and create charts.
30 core toolsRun scheduled pipelines with conditions, approvals, subtasks, webhooks, notifications, model/tool overrides, safety modes, run history, and insight cards.
AutomationTelegram, WhatsApp, Discord, Slack, and SMS share streaming, media intake, reactions, approvals, health checks, local speech-to-text, and Kokoro TTS.
5 channelsInstall sandboxed plugins, connect external MCP servers, coordinate Claude Code CLI through an approval-gated skill, and migrate selected Hermes/OpenClaw data with backups.
ExtensibleGenerate or edit images through OpenAI, Google, and xAI, then create video clips through Google Veo or xAI Grok Imagine Video from the same provider-aware model catalog.
Provider mediaOpenAI, Anthropic, Google AI, xAI, OpenRouter, MiniMax, custom endpoints, local Ollama rows, and ChatGPT / Codex all flow through one catalog with provider-aware labels.
Updated in 3.20Thoth remembers people, projects, preferences, facts, events, and relationships as a graph, then retrieves relevant nodes and one-hop context before each response.
Document extraction, Dream Cycle refinement, provider-aware insight snapshots, wiki export, and the interactive graph view keep that memory useful instead of turning it into a pile of notes.
One assistant, local models, API providers, custom endpoints, MiniMax, and ChatGPT / Codex
Run inference on your hardware with 39 curated models across Qwen, Llama, Gemma, Mistral, DeepSeek, Phi, and more.
qwen3:14bAdd OpenAI, Anthropic, Google AI, xAI, OpenRouter, MiniMax, custom OpenAI-compatible endpoints, or ChatGPT / Codex when you need frontier reasoning or do not have a GPU.
Ask a question in natural language and the agent autonomously decides which tools to chain together. Search the web, check your email, run a shell command, inspect provider and insight status, build a deck in Designer, and chart the results - all in a single conversation turn.
Connected channels add send tools automatically, external MCP servers add dynamic tools, and every risky action stays behind Thoth's approval gates.
Core tools ship with the app; channels and MCP servers extend the tool belt when enabled
Configure provider accounts, pin model Quick Choices, build presentations and interactive pages, automate recurring work, route approvals to messaging channels, and connect MCP servers without making startup fragile.
ChatGPT / Codex sign-in is separate from OpenAI API keys, Claude Code Delegation stays approval-gated, and imports from Hermes/OpenClaw remain scan-first with backups and redacted reports.
Cloud AI rents you access. Thoth gives you ownership.
| Cloud AI | 𓁟 Thoth | |
|---|---|---|
| Your data | Stays on your machine - always. Provider models receive only the current conversation plus model-visible tool context | |
| Conversations | Local SQLite, fully yours, exportable anytime | |
| Cost | Free with local models. Providers use upstream API billing or your ChatGPT subscription | |
| Memory | Knowledge graph - 10 entity types, 67 relation types, visual explorer, FAISS semantic search | |
| Tools | 30 core tools + channels + plugins + MCP tools | |
| Customisation | Swap models per thread, mix local + providers, configure identity, enable skills and guides, build workflows, install plugins | |
| Voice | Local Whisper STT + Kokoro TTS - never leaves your mic | |
| Messaging | 5 channels - Telegram, WhatsApp, Discord, Slack, SMS - with streaming and media | |
| Availability | Local models work fully offline |
Use the one-click Windows installer, macOS DMG, or verified Linux one-line installer. Each release ships self-contained runtime assets for the supported platform.
First-launch wizard: Local - download an Ollama model (default: qwen3:14b), or Providers - enter an API key and seed Quick Choices. ChatGPT / Codex sign-in lives in Settings after launch.
Ask anything in natural language - the agent handles the rest. Search the web, check email, analyze your screen, track a medication, schedule a workflow.
Installs the latest verified XDG user tarball, creates ~/.local/bin/thoth, and uses the same release manifest as in-app updates. Manual tarballs remain available from GitHub Releases.
curl -fsSL https://raw.githubusercontent.com/siddsachar/Thoth/main/installer/install-linux.sh | bash
Thoth's app has no account system, server, or telemetry pipeline. Conversations, memory, Designer projects, documents, and local model runs stay on your machine. Provider models, ChatGPT / Codex, and online tools are opt-in, with secrets stored in the OS credential store when available.
OLLAMA_HOST ports, URL forms, wildcard binds, local vision catalog rows, and busy checks now resolve through the normalized local endpoint.8080 is really Thoth before reusing it, moves to the next free port when needed, and keeps Windows/macOS tray behavior intact while Linux defaults to browser/no-tray.Free, open source, and yours to keep. No Thoth account. No Thoth subscription. No data harvesting.