Amadeus
A desktop AI agent combines voice chat, optional characters and delegated tasks, with a local app, local voice models and remote DeepSeek chat.
Amadeus is a desktop AI agent for people who want voice conversations and delegated work in one interface. It keeps task progress, permission requests and results visible while agents work.
- Voice conversation: Speak or type, and interrupt spoken responses while they're playing; local speech recognition uses Qwen3-ASR or SenseVoice, with GPT-SoVITS for speech synthesis.
- Task delegation: Use Pi for everyday tasks, Codex for coding, a Browser provider for managed pages, or optional OpenClaw integration.
- Work oversight: Review structured diffs and artifacts, approve actions, retry attempts and recover work after a restart; actions require individual approval by default.
- Optional characters: Character packs coordinate subtitles, lip sync and expressions with speech, while chat and work also function without a character.
The desktop app and its backend run locally on Windows, macOS and Linux. Main Chat defaults to remote DeepSeek, so that configuration sends chat requests to an external service and needs internet access. An optional llama.cpp profile provides local LLM inference. Local voice failures don't silently send audio to a remote service. MCP tools and Skills work through compatible task providers, rather than directly through Main Chat.
Basic text chat and remote voice need no GPU. The validated Windows local-voice setup targets an NVIDIA GPU with 8 GiB VRAM and at least 16 GiB RAM; a local LLM needs additional memory. CPU local voice is available without a real-time performance guarantee. Intel Macs support basic chat and remote voice but can't use the local-model profiles.
Amadeus is open source under AGPL-3.0 and comes as runnable source rather than a self-contained desktop release. Character packs, model weights and reference audio aren't included; external assets retain their own licenses.