FlowKy assumes you already have an agent and pay for it. It fits around that loop rather than replacing it.

The loop it is built for

You see something wrong in a browser. You capture it, with a comment. That capture reaches your editor with the console output, the failing requests and the screenshot attached — so the agent starts from evidence rather than your description of it.

The agent changes code. You replay the flow that covers the area it touched. If the flow still passes, the change did not break the path that mattered.

What the agent gets that it would otherwise guess

The current state of the page, not a description of it. The actual console errors. Which requests failed. The flows that constrain the area. The tasks already filed against the site. The brand details so it stops inventing placeholder copy.

What it does not do

It does not run your agent, choose your model, or write prompts for you. The AI features are on your own key, and there are none that run without you asking.

Which agent

Any of them. The bridge and the context are agent-agnostic — Claude Code, Cursor, Copilot and Gemini all read the same repository.