How Denker Works
Denker is a new interface for AI agents. Instead of typing into a chat window, you point, draw, or talk on your screen: Denker sees what you mean, dispatches the work to your agents — running on your own Claude Code, Codex, or OpenCode — and the result comes back on your screen where you can keep working with it.
The model
Hand off, agents work, result returns
Every interaction with Denker follows the same loop. You never leave your app to explain work to AI: Denker sees what is in front of you, so a short instruction plus a point at the screen is enough to hand work over.
From there your agent team takes over. Agents work in the tools you connected — researching in the browser, drafting replies in Gmail, updating a spreadsheet — and hand follow-up steps to each other until the task is done.
Then the result comes back to you: a frame on your canvas when you need to review it, or written directly into the app where the work belongs. You stay in control throughout — redirect agents, reprioritize, or edit their work, without babysitting.
The workspace
One canvas for the whole team
Humans and agents share one infinite canvas. Everything lives there — research, code, emails, data — each in its own frame, rendered live as agents work. No tabs, no hidden windows, no buried chat history: your entire workspace at a glance.
On the Mac app your agents are also present on your screen, each with its own cursor that moves with intent while it works alongside yours. You see what they are doing in real time, on top of whatever app you are in.
Core concept
The Taskboard
At the center of every project is a Kanban board on your canvas. Give Denker a task and it breaks it into steps:
Context
Memory: things and facts you can inspect
Memory connects people, projects, apps, and other things with the facts Denker has learned. Select a thing on the globe to inspect its details and linked facts inside the frame. Review suggestions before accepting them, so useful context stays clear and inspectable. Memory is part of Pro, including the 14-day Pro trial.
Agents pick up tasks on their own
Your agents don't wait for your next message. They look at the Taskboard, pick up the next task, execute it, and move it to Done. Each agent runs your choice of AI CLI — Claude Code, Codex, or OpenCode — so you can pick the best model for each job and mix and match across the team.
While one agent researches, another writes, another codes. Every agent has its own cursor on the canvas. Every output is a frame you can inspect.
You stay in control — redirect agents, reprioritize, or edit their work. But you don't have to babysit.
Everything is a frame
Every piece of agent output appears as a frame on your canvas — research summaries, code, emails, search results, workflows. Move them, resize them, organize them. No hidden state, no buried chat history.