What is Celeris
Celeris is an app for getting small tasks done. Press a global hotkey, type one line of intent ("raise a PR", "reply to this", "book this in"), and Celeris works out what your screen implies and does it. It asks first if anything is about to change.
It runs on celeris-1, a diffusion model built for speed. Small tasks finish while you are still looking at them.
Built for a fast model
A harness built for a slow model spends its time on long chains of small model calls: one narrow tool per operation, and a plan worked out again every turn. A step that works nine times in ten works about half the time when you need six in a row. Celeris is built the other way round:
- Tools do whole tasks. "Raise a pull request" is one tool that drives git and the GitHub CLI, not six steps to sequence correctly. One decision and one approval instead of six.
- A step budget. A turn that is not converging stops rather than running on.
- The screen is the context. The hotkey captures what you are looking at, so you rarely have to explain where you are.
What it does
- Chat, from a hotkey over whatever is on screen or from the chat window.
- Tools, connectors and plugins: built-in tools, the services you already use, MCP servers, and bundles of skills you install in one step. Celeris learns which tools work for your tasks.
- Triggers save a repeated task behind one key or a schedule.
- Your phone chats with celeris-1 directly, or sends work to your computer and watches it run there.
- Voice: dictate a task and hear the answer read back.
Reads run on their own; anything that writes, sends or deletes stops and asks. See Approvals & safety.
Where it runs
The desktop app runs on macOS, Windows and Linux, on your own machine. Your screen captures, files and command output stay there, apart from what is sent to the model to answer the turn you asked for. Privacy lists exactly what leaves the device.
Celeris is free to install. Sign in with a Celeris account and choose the workspace its model usage is billed to. See Pricing.