Parallel agents

Denker splits a big task into independent parts and runs helper agents on them at the same time, so you follow one request instead of many.

Updated

Summary

Some requests hold several jobs that don't depend on each other: research three competitors, check five pages, draft replies to a few emails. Denker hands those parts to helper agents from the AI tool you already use, Claude Code, Codex or OpenCode, and they work at the same time. Denker picks a fitting model for each helper. You keep talking to Denker; each helper shows up as its own card on the Taskboard, and the results come back to your chat.

Key points

  • Ask once. Denker decides which parts can run side by side and starts a helper for each.
  • Watch it on the board. Each helper is its own card with an arrow and its parent's number, and the parent card counts its helpers by status.
  • Steer while it runs. Send a follow-up while Denker works and it changes the running task; stop or resume an agent from its task.

Give Denker the whole job

Describe the outcome and the parts in one request, for example: “Compare the pricing pages of these three competitors and summarize the differences.” You don't assign agents or pick models; Denker splits the work where parts are independent and keeps parts that depend on each other in order.

Helpers run on the provider's own sub-agents in your AI tool, so they use the same sign-in as your main chat. What a helper can reach depends on your setup and connected apps.

  1. Ask for the whole result in one chat message or by voice.
  2. Denker starts a helper for each independent part.
  3. Follow the helpers on the Taskboard or in the work toasts.
  4. Read the combined answer in your chat and open any helper's card for its details.

See every helper on the Taskboard

Each helper a run starts appears as a subtask card with an arrow and its parent's number on top. The parent card shows how many of its helpers are running, waiting or done, and you can expand it to see them. The count updates as the helpers finish.

Work toasts follow each task live while it runs, including questions an agent needs you to answer.

Run more than one chat at once

Parallel work also means separate chats. Start a new chat for an unrelated request while another one is still working; the Chats tile groups them as Running, Today and Earlier, and you can stop, rename or archive each one.

Review the combined result before you use it. Sending, publishing and payments stay explicit steps for you.