Interactive panes

How to use Agent View panes.

Visible panes are for work you want to watch and steer by hand: review, side-by-side provider comparisons, standby agents, and manual driving. This is how to open a pane, choose its model, and run your prompt across providers.

At a glance

  • Agent View shows live, interactive panes for work you want to watch and steer instead of sending to the background.
  • Use it for review, provider comparison, standby agents, and manual driving.
  • Each pane runs one CLI session against one provider and model — Claude, Codex, Gemini, or your local model.
  • Open as many panes as you need; the grid auto-lays them out for you.
agent view — four panes
CrazyToad Agent View with four visible panes
Note
Spawn as many as you need — the grid auto-lays them out. Agent View is an auto-arranged grid, not a free drag-and-resize canvas.

Open as many panes as you need

Ask Toad with watch-and-drive language like spawn here, give me a pane, or let me drive. That intent puts a visible, interactive pane in Agent View instead of sending the work to a background agent. By default, delegated work goes to background agents; the watch-and-drive vocabulary is what routes it into a pane you can see. Open as many panes as you need — the grid auto-lays them out for you, so you never drag or resize to fit them.

Choose the provider and model

Each pane runs one CLI session against one model. Pick the provider per pane — Claude, Codex, and Gemini ship as defaults, plus your own local model once you add it in Accounts. CrazyToad runs on almost any model, brain and agents alike. (The brain itself runs on almost any model, though not the Gemini or Antigravity CLI — those stay agents-only.)

Run the same prompt across providers

Open one pane per provider and send the identical prompt to each. Because every pane shows its live session, you can compare how Claude, Codex, Gemini, and a local model approach the same task side by side. When you would rather hand the whole thing to Toad and let it route the work, use Studio instead — Toad decides how to split and run it.

Keep a standby pane ready

Leave a pane idle as a standby agent for quick manual jobs — a one-off command, a fast lookup, or a sanity check — without spinning up a whole background task. The pane stays put on the grid, ready the moment you need it.

Steer and review the live output

Because the agent is visible, you stay in control: nudge it mid-run, inspect what it is doing, or stop the work before it goes further. Review the result right in the pane, then promote it, discard it, or send a follow-up.

Four ways to use a pane

Agent View suits any task where watching beats walking away. Reach for these patterns.

ReviewOpen a pane to read an agent's work as it happens and catch a wrong turn early.
Provider comparisonRun one prompt across Claude, Codex, Gemini, and local models in parallel panes, then compare.
Standby agentKeep an idle pane on hand for quick manual jobs without queuing a background task.
Manual steeringDrive the session yourself — nudge, redirect, or stop the run from the live pane.
Related