Changelog
What changed, newest first. Early access: things move quickly.
Talk instead of typing
Dictate in the message box with the microphone, or with Wispr Flow, Superwhisper or your system dictation: long and multi-line text lands intact and Enter never sends mid-dictation. Hands-free conversation reads answers aloud and you can interrupt by speaking. In the terminal, omni voice and /voice dictate, and a pasted or dictated block stays one message. Audio is not stored; the page names who hears it.
Headless mode, SDK, docs and status page
omni -p answers one question on stdout and reads piped input, so it works in scripts and CI. A small typed JavaScript SDK talks to the same chat-completions compatible API. Docs, this changelog and a status page are now part of the site.
Run a change on your own computer from the chat
The chat can send a change to a paired computer (the relay), show the same live run view, and ask you questions mid-run. Agent tools gained to-do lists, MCP tools and a safer shell with controlled network access.
Computer use, research and voice
A guarded computer-control session, a research mode with checked citations, push-to-talk dictation and an installable app that works offline for reading.
Conversation history that follows you
Optional, off by default: server-stored history with search, export (JSON and Markdown) and import. Conversations stay in your browser unless you turn it on.
Hardening and speed
A security pass over the API, rate limits and idempotency on writes, a performance budget and fixes from end-to-end runs on a clean machine.