Sirius is the first ambient intelligence for macOS.
Hold fn and speak, or press ⌘⇧J to type, and your agent gets to work in its own folder. Replies come back as a toast, a notification or a spoken line.
We believe the future is multi-agent, so Sirius is bring your own harness (BYOH): it has no model of its own and drives the agent CLI you already use, with your account and settings. Claude Code, ChatGPT / Codex, Cursor, Google's Antigravity (Gemini) and Muse Code work out of the box; others are one adapter away (docs/byoh.md).
Download Sirius for macOS 14+ on Apple Silicon.
| Shortcut | What it does |
|---|---|
Hold fn and speak |
Send to an agent. Start with its name ("Claude, …") to pick one |
Hold ⌃ and fn |
Paste what you said at the cursor, in any app |
| ⌘⇧J | Open the composer; fn dictates into it |
| ⌘/ (in the composer) | Switch to the next agent |
Change both shortcuts in Settings…. Speech recognition runs on your Mac; the first start downloads the speech model (about 145 MB), or reuses Parrot's model and dictionary. The full manual ships in the app under Sirius Manual, and your agent can answer questions from it (read it here).
Each agent is a CLI session Sirius keeps warm in its workspace through a per-harness adapter. The agent reaches you by calling Sirius (the sirius command or the sirius-mcp server), never by its own output. See docs/architecture.md and docs/byoh.md.
swift build && swift test
scripts/dev-install.sh # build, sign, and install Sirius.appWithout a Developer ID in your keychain the app is ad-hoc signed, and macOS forgets its permissions on every rebuild.
Apache-2.0. See LICENSE and NOTICE. Report bugs at github.com/humanitas-labs/sirius/issues with ~/Library/Logs/Sirius/sirius.log attached; it records what Sirius did, not what you said.