Agents & Ask Deska

It helps to keep the two apart:

• Ask Deska (Cmd/CtrlT) is a quick chat bar that drops down over the canvas. It can see what is open — your panels, running agents and terminal status — and can act on the canvas: open or close panels, run a command, read a panel, or hand work off to an agent.
• The Agent panel is a full coding agent living in its own panel, with a thread, live tool calls, a model picker, adjustable reasoning, and saved sessions you can resume.

A rule of thumb: reach for Ask Deska to ask about or rearrange your workspace and kick off quick tasks, and open an Agent panel when you want a focused, multi-step coding session with full control.

Press Cmd/CtrlT from anywhere to open Ask Deska. Type a question and it streams back an answer, using the current canvas as context. What makes it different from a generic chatbot is that it can do things, running small tools inline as it works:

• See the canvas — list your open panels, terminals (idle vs. running, which ports are listening, which agents are detected) and agents, or pull together the recent errors across all of them.
• Read a panel — pull the live contents of a terminal’s scrollback, a browser’s page, an editor’s file, an agent’s transcript, or a Git panel’s branch and diff summary.
• Move things into view — focus a single panel or highlight several to walk you through them.
• Open and close panels — spawn a browser at a URL, an editor on a file, a terminal that runs a command, or an agent that starts on a task — and close panels when they are done.
• Drive existing panels — type a command into a terminal that is already open, or hand a follow-up task to a running agent.

• @-mention panels to point it at specific tabs — type @ and pick a panel to pin it as context for your message.
• Scope the context to just the current workspace or all of them with the scope chip, so you control how much it ships with each message.
• Pick the provider and model from the buttons in the bar.

Open a Deska Agent from the Panel Library (Cmd/CtrlShiftE). It gives you a chat thread that can read files and understand your workspace, with live tool execution, a sidebar of recent sessions, and a header that holds the model picker and a stop button. You can switch models mid-conversation.

A few controls shape how it works:

• Reasoning effort — dial extended thinking from off up through minimal, low, medium, high and extra-high (shown as filling bars). Higher levels make the model reason longer before acting, at the cost of more time and credits.
• Tool approvals — when the agent wants to run a command or edit a file, it pauses and shows an Allow / Deny card right in the thread. You can flip on auto-approve for a panel (offered when you create the agent) to let it run without stopping to ask.
• Steering — choose whether the agent executes its planned tool calls all at once or one at a time, so you can keep a tighter rein when you want to.
• Auto-compaction — as a long conversation approaches the model’s context limit, the agent automatically summarizes earlier turns so the thread can keep going without losing the thread of what you are doing.

Agent conversations are saved to disk and listed in the panel’s sidebar for the current project, so picking up where you left off is a click away — and they resume automatically when you reopen the workspace.

A stats chip in the header keeps the numbers honest: how much of the context window is in use, the input, output and cached tokens billed so far, and the running total — in credits on managed Deska AI, or in dollars when you bring your own key.

You are not limited to Deska’s own agent. You can launch popular standalone agent CLIs straight into a terminal from the Panel Library — Claude Code, Codex and OpenCode. The Library checks whether each tool is installed and offers to install it for you, and an auto-approve toggle lets you skip its permission prompts.

Deska also resumes those conversations after a restart — Claude Code, Cursor, Gemini CLI, Codex and OpenCode all pick the thread back up. Each CLI handles its own sign-in and models.

Inference reaches a model one of two ways:

• Deska AI (managed) — you sign in and Deska runs the model on its own keys. There is nothing to configure; usage draws from your monthly credits. This is how Free and subscription plans work.
• Bring your own key (BYOK) — you supply your own API keys for Anthropic (Claude), OpenAI, Google Gemini, Groq or OpenRouter. Inference then runs on your account and your budget, never touching Deska’s credits.

Both surfaces let you pick a model from the active provider. Managed Deska AI sorts its catalog into two tiers:

• Standard models are fast and inexpensive — good for everyday questions and most edits. They are available on every plan.
• Premium (frontier) models reason harder and cost more credits per message. They unlock on the higher subscription scales (and always on Lifetime).

Many models support extended thinking, which you can dial from off up to extra-high when you want the model to reason harder. Credit costs scale with the real cost of each request, so a bigger or smarter request spends more — the exact, current rates per model live on the pricing page.

The default shortcut for Ask Deska is Cmd/CtrlT — and you can rebind it under Settings → Keybindings. (That is also why a new terminal lives on Cmd/CtrlShiftT: the plain chord goes to the assistant.) You can also fire a natural-language request at an agent straight from the Command Palette.