Amadeus
An open-source desktop agent that combines a voice character with real coding-agent-style work, letting a user speak or type naturally while specialized providers execute delegated tasks and the character narrates progress and asks for permission.
Amadeus brings a spoken conversation, an animated character with lip sync and expression, and delegated task execution into one interface instead of three separate windows for chatting, performing and running a terminal. The character communicates and narrates while specialized Work Providers actually execute tasks, and a Host component owns identity, state, permissions, persistence and recovery, so a long-running task stays visible with progress, permission requests and the ability to resume or take over.
The desktop application and its local Host run on the user's machine, with either remote model services or local inference selected through configuration; character packs that add voice and visual presence are optional, and chat and delegated work both remain available without them. Local model inference is supported on NVIDIA CUDA, with experimental support on AMD ROCm and Windows, and on Apple's MPS backend. The project states plainly that this branch is buildable, runnable source for an early 0.15.2 alpha rather than a packaged release, and that first-party code is licensed under AGPL-3.0 while bundled third-party code and assets keep their own terms.