Where it runs
Work starts where you are, runs on machines you choose, and thinks with models you choose. flower is built in three layers: what you drive it from, where the models come from, and where the work runs. Each layer can be yours. Here is what holds today, and the one thing still to come.
01 Drive
Where you drive it from.
You plan, steer and read the work from wherever you already are. Both of these reach the same flower, and you can use either at any time.
Now
- Your own terminal agents, Claude Code, Codex or Pi, connected over MCP. flower is a server they ask; there is nothing new to learn inside the agent.
- flower’s web UI in any browser, on a desktop or a phone, from anywhere you have a connection.
02 Models
Where the models come from.
Models are not tied to the machine the work runs on, and you choose per kind of work: one model for one job, another for the next.
Now
- Your own subscriptions, so a seat you already pay for does the work.
- Provider APIs, when you would rather pay for what you use.
- Self-hosted or cloud GPUs, running open models.
03 Compute
Where the work runs.
A build works in a contained workspace of its own, so one build cannot trip over another, and nothing reaches your main branch until it has been reviewed and checked.
Now
- Contained workspaces on flower-run machines, in the cloud or on bare metal.
Coming · talk to us
Bring your own compute.
Your own servers, in the cloud or on your own premises, running the same contained workspaces where your data and your networks already are. It is not self-serve yet: hello@useflower.dev.
From almost fully cloud to air-gapped.
flower is built so every layer can be yours: from almost fully cloud to running on your own hardware with your own models, air-gapped. That is a design property and the direction the whole thing points, not a switch you flip today; each layer moves to your side as its piece is ready.
In between, the layers mix: your agents driving flower’s machines, flower’s web UI over your own models, one layer on your side while the others are not yet.
How it works, from a session to a live deploy, or what flower connects to in its own words.