How Denker Works
Start with the email, page or document you are already working on. Point or draw to show the relevant detail, then explain the result you want. Denker carries that request to your configured AI and gives you a place to follow the work and continue from its output.
Summary
Denker connects a visible starting point with an agent task and a result you can use. For a customer email, show the message, ask for a reply confirming the next steps and review the draft. Tasks, conversations and workspace frames help keep that exchange together when you need a correction or another step. Available actions depend on your configured AI, app access and task.
Key points
- Show the work you mean. A point or marked view identifies the email, paragraph or interface detail. Your words describe what should happen, so the agent has a clearer starting request.
- Keep the task connected to its progress. Follow the conversation and task state, provide missing context and clarify the next step. A longer request has somewhere to continue after the first response.
- Continue from a useful result. Review the draft, changed app or workspace frame where it belongs. You can refine it with the agent and decide when it is ready for use.
The model
Example: turn a customer email into a reply draft
Open the email you need to answer. Mark the relevant message and ask Denker to draft a reply confirming the next steps. The marked view identifies the source, while your instruction gives the reply its purpose. Include any facts the visible message does not contain.
Your configured Claude Code, Codex or OpenCode agent works with the information and tools available to it. If the reply needs a current project document, provide that document or use a supported connection that can read it. Follow the conversation and answer any question that blocks the draft.
Review the returned draft against the email and your instructions. Ask for a correction if a fact or commitment is wrong. Where supported, ask Denker to put the reviewed text into the reply field, then check it there before sending. The original request and follow-up give the work a thread you can continue.
Watch Denker turn a Google Sheets report into a draft Gmail reply. This edited recording uses fictional data. The reply explains the figures, flags missing information and stays unsent for review.
Write with your voice
Dictation with the finish you choose
Dictation anywhere turns spoken words into a draft for the app you are using. An optional prompt can make the text concise, organize it into bullet points, or translate it. Draw while dictating to include marked screenshots in apps that accept image pastes. Review the text and visuals together before sending.
Dictation setup, visual references and prompt examples →The workspace
One canvas for the whole team
The canvas gives related material and agent outputs a place together. Keep a draft beside the notes you used to prepare it, or arrange a research summary next to the task it supports. Move and resize workspace frames to make the current work easier to find.
During supported on-screen work in the Mac app, Denker can show an agent cursor or highlight the relevant area. These cues help you follow the target while the task develops. Use the task and conversation to understand what needs your attention.
Core concept
Find the next step on the Taskboard
Use the Taskboard to see which work is waiting, blocked or ready for review. Open a task to find its request, assignee and ongoing conversation. For the customer reply, that keeps the draft and any unresolved question connected to the work you asked for.
Context
Reuse useful context with Memory
Company details or a project decision can matter to more than one task. Memory keeps learned facts linked to the people, companies and projects they concern, so relevant context can be reused. Inspect a saved detail on the graph and correct it when it changes. Review proposed changes where offered. Memory is part of Pro, including the 14-day Pro trial.
Work in your apps
Computer use
Ask Denker to apply information in the app where you need it. For a supplier form, relevant saved company details can reduce repeated typing. Show the fields, identify the facts to use and ask it to stop before submission. This Memory example requires Pro or its trial and supported app access. Review the filled fields before continuing.
Computer use setup, task boundaries and result review →Divide work and keep contributions connected
A larger request may benefit from separate assignments. One agent can collect product sources while another prepares a draft from those findings. Give each task an outcome and identify which result the next task needs. Use the Taskboard to follow those contributions in the same project.
Available agents use your configured Claude Code, Codex or OpenCode setup. Team controls and task execution depend on that setup and the tools each assignment needs. Start with one agent and add another when a separate responsibility helps the work.
Review the contributions before depending on them. Clarify a blocked question, change a task's priority or ask for a correction against its result. That keeps the team focused on the outcome you want.
Keep useful results beside their sources
Workspace frames can hold research summaries, drafts and other returned material. Arrange them around the source you are reviewing so a correction has a clear reference. When the work continues in a coding tool or another app, use its linked session or actual destination to inspect and continue that result.