How to drive Toad in Studio.
Studio is the default surface: one conversation with the brain, with the work docked around it. This guide walks the loop — ask Toad for work, choose where it runs, approve decisions inline, and pull proof up beside the chat.
At a glance
- Studio is the default CrazyToad surface — one conversation with the brain, with the work docked around it.
- Toad decides how to run each request: answer directly, delegate to background agents, or fan out a bigger build.
- Approvals — plans, memory, code, docs — surface inline in the conversation, not in a separate queue.
- Open Workbench panels beside the chat to inspect the same evidence Toad sees.
- Sessions resume per workspace: switch repos and pick up where you left off.
Type the work in the composer
Describe what you want in plain language — a one-line fix, a whole feature, or a question about the codebase. Toad reads the request, plans the work, and either answers directly or delegates it to agents. Keep the conversation going: follow-ups stay in the same thread, so Toad keeps the context instead of starting over.
The brain that reads your request runs on almost any model you wire up in Accounts — a Claude or Codex subscription CLI, a raw API key, or a local model. The one exception is the Gemini/Antigravity CLI, which is an agents-only lane and cannot host the brain.
Choose where the work runs
Every request lands somewhere. Toad decides the default routing, but you can steer it with the words you use.
Approve plans, memory, and code inline
When Toad needs a decision — a plan to run, a memory to save, code to merge, a doc to change — the approval shows up right in the conversation. Read it in context and approve or redirect there, instead of hunting through a separate queue. Memory in particular is never written silently: it is evidence-linked and only saved when you approve it.
Open Workbench panels for proof
When you want to see the result, open a panel from the Workbench dock — Preview for a rendered page, Files and Git for the source, Terminal for a focused check. It docks beside the chat so you and Toad inspect the same evidence. Git integration runs through the gh CLI, so repo and pull-request actions use the same GitHub login you already have.
Resume across workspaces
Switch workspaces to move between repos. Each switch reloads that workspace's session state and brain memory, so you pick up where you left off instead of rebuilding context from scratch. One workspace is active at a time; the rest keep their own history until you return.
Talk with voice when it is faster
Voice and live-voice controls sit next to the composer, including the xAI realtime lane. Use them for hands-free intent capture when saying the request is faster than typing it — Toad transcribes what you said and drives the same chat loop.