Screen group

AI & agentic

Streaming answers, agent runs, approvals, and the failure modes that only appear once a model is doing the work.

AI & agentic · 7 available

Available ai & agentic screens

Live routes rendered in isolated frames. Each preview has its own screen-size control — or open a route in a dedicated tab.

Tokens arriving into a settling layout, with stop, regenerate, and a partial answer that stays readable.

Steps expanding as they execute — tool calls, inputs, results, nested subtasks — while the run is still going.

A proposed action shown with its full effect, and approve, edit, or reject before anything leaves the app.

Inline reference marks tied to a sources panel, so the claim and its evidence stay visible together.

Chat on one side, the thing being built on the other, with version history between them.

Refusal, context limit, tool error, cancelled run, and exhausted credits as distinct, actionable surfaces.

Ghost text, a field-level suggestion, and a summarize action living inside an otherwise ordinary form.