Sutando is an open-source AI agent developed by sonichi, designed to run natively on macOS. The project emphasizes real-time operation and a self-improving architecture, allowing the agent to learn from user interactions and autonomously update its own code. It supports various modalities including voice, vision, and screen interaction, enabling it to perform tasks such as rewriting text on screen, joining meetings, making phone calls, and working from a phone via screen sharing.
The agent is designed to integrate with common communication platforms like Zoom, Google Meet, Telegram, and Discord. A key feature is its ability to operate using an existing Claude Code subscription, aiming to minimize additional API costs often associated with other AI agents. Sutando is described as an 'Alpha' stage project, with a demo showcasing its ability to perform multiple tool calls and tasks by voice from a phone. The project also highlights its capability to scale across multiple Mac machines, with agents coordinating handoffs without manual migration scripts.
Sources
Claim check: 5/5 supported claims - 5 evidence links - 94% avg confidence
- Sutando is an open-source AI agent for macOS that operates in real-time and can rewrite its own code.supported - github.com
- The project has 335 stars and 66 forks on GitHub.supported - github.com
- Sutando's latest release is v0.1-demo, published on April 13, 2026.supported - github.com
- Sutando can perform 24 tool calls and 6 tasks in 7 minutes by voice from a phone, as demonstrated in a video.supported - github.com
- Sutando runs on existing Claude Code subscriptions and aims to minimize extra API costs.supported - github.com
Caveats
- The self-rewriting capability is a stated goal and feature, but its extent and effectiveness are not independently verified beyond the project description.
- This is based on a demo video linked in the GitHub release notes; independent verification of the performance metrics is not provided.
- This claim is from the project's README and has not been independently verified for cost efficiency.
- Single-source caution: verify critical details at the linked source.
Radar score 82/100 - how it was calculated
- Reliability 82: GitHub metadata supports source trust
- Freshness 92: Fresh GitHub activity
- Novelty 66: Novelty blends source metadata and enrichment
- Technical 74: Repository technical metadata
- Developer 95: Developer tooling signals
- Ecosystem 61: Developer-oriented GitHub signal
- Confidence 96: Claims have reliable evidence
Discussion
Loading comments...