Run FlowAI's agents on your own machine.
FlowAI Desktop runs a full software-delivery workflow — architecture, implementation, peer review and validation — on your laptop, using your own provider keys and your own git credentials. FlowAI's servers plan the work and decide the gates; your code and your keys never leave the machine.
Downloads are not published yet — create an account and we will tell you when they are.
macOS 11 or later, or Windows 10 or later. Bring your own key from Anthropic, OpenAI, DeepSeek or Qwen.
FlowAI
Your machine
Your code never leaves it.
The same governed workflow, executed locally.
This is not a cut-down offline mode. A desktop run is planned, validated and gated by exactly the same server-side workflow as a cloud run — the difference is where the work happens. FlowAI decides what to do next and whether the result passes; your machine holds the repository, calls your model provider, and runs the coding agent.
Your machine does the work. FlowAI keeps it honest.
Your keys, your code
The coding agent runs with the provider key stored in your OS keychain, inside a workspace folder you choose. FlowAI's servers never receive the key and never hold a copy of the workspace.
- Anthropic, OpenAI, DeepSeek or Qwen — whichever you already pay for
- Keys are stored by the operating system, not by FlowAI
- The server sees run status and token counts, never source
The whole delivery chain
Not just code generation. Every stage of the governed workflow runs on your machine, in order, and a stage that fails validation goes back for remediation rather than through.
- Architecture design, implementation, then peer review by a different engineer
- Structural, engineering-quality and grounded-truth validation
- A final delivery gate that can still send the work back
Git on your terms
Cloning, branching, committing and opening a pull request all happen locally with your own credentials — and only while you have Commit & PR switched on.
- Feature branches and pull requests on GitHub or GitLab
- Switch Commit & PR off and nothing is written to a remote
- No git on the machine? The run still completes; it just skips git
Review and control
A local run is a first-class run: it appears in your workspace alongside cloud runs, with the same controls and the same audit trail.
- Pause for human review, then approve or reject and resume
- Cancel a run, or rerun a single stage that already finished
- Work returns to the machine that holds its workspace
Three steps to your first local run.
01
Install FlowAI Desktop and sign in with your FlowAI account.
02
Add a provider key and choose a chat model. FlowAI stores it in your OS keychain, never on our servers.
03
Describe a task in chat or on the Tasks page. It runs on this machine, and the deliverables land in your workspace folder.
The local agent ships inside the app and starts with it — there is nothing separate to install. If you prefer a headless machine, the same agent is available as a command-line tool.
Start in the browser. Move to your machine when you want to.
One account covers both. Runs you start in the web app and runs you start on your laptop share the same teams, the same history and the same review queue.