Hephaistos
A local-first, multi-device personal AI gateway
Hephaistos brings conversation, memory, research, documents, voice and automations into one interface, with a clear choice between optional hosted Cloud and a compatible personal server.
The need
AI usage quickly becomes fragmented across conversations, documents and tools. Hephaistos aims to keep personal context coherent while letting users choose where their assistant runs.
The experience
The mobile app offers streaming Jarvis chat, preference memory, sourced web research and file analysis in an interface designed for everyday use.
- Voice, attachments, favourites and export
- Agents and scheduled automations
- Multi-device sync and diagnostics
- Agora community and client-side encrypted direct messages
The architecture
The Flutter client talks to Supabase and a FastAPI backend over HTTPS and WebSocket. On a personal Apple Silicon server, inference uses MLX and Metal. An encrypted local outbox protects actions waiting for acknowledgement.
Production status
The public iPhone and iPad version is available on the App Store. The source also targets Android, but no public Google Play availability is claimed until release is verified.
Architecture
- Client
- Flutter · Riverpod · GoRouter
- API
- FastAPI · Python · WebSocket
- Local AI
- MLX · mlx-vlm · Metal
- Data
- Supabase Auth · PostgreSQL · Realtime
Constraints stated transparently
- A compatible server or Hephaistos Cloud is required to use Jarvis.
- The production personal-server runtime currently focuses on macOS Apple Silicon.
- Jarvis conversations are not presented as end-to-end encrypted.
- Android is in preparation with no verified public store link.