Skip to content
Agent Process

Processes your agent follows

A new conversation shouldn’t mean explaining your sources, your steps and when to ask you all over again. Turn your routine into a process your agent follows, with the results it must hand in and the points where you review.

Build with your agent · Refine on the canvas · Run from your agent

Process canvas / Research brief

A clear path from question to reviewed work

Published · v1

Start · Choosing an AI coding tool

Needs another pass? Return to the brief.

Agent workYour checkpointClick a step to inspect it

Step 1 / Requirements

Gather sources

Input: your topic. Required output: source notes with links.

Owner: Your agent

Ready to run. Start with a topic, finish with a reviewed brief.

Interactive illustration · Run the example to try required evidence and rework. No live systems connected.

Tell your agent how the work goes.

Your source list. Your comparison criteria. Your review before the brief is final. Ask your agent to help build that routine, then refine the same PROCESS.md as text or on the process map.

Describe it to your agent

“Use my source list. Compare the options on cost and privacy. Attach the links, then ask me to review before it’s final.”

Illustrative instruction · The text and the map are two views of the same file.

Your requirements / Write the comparison

Owner
Your connected agent
Inputs
Your topic + chosen sources + cost and privacy criteria
Done means
A comparison brief and its source links are submitted
Next
You review → approve or ask for another pass
Shape the flow around the way your work unfolds

Choose the route

Name the routes a step can take. The agent or person doing it picks one and gives a reason, and both are recorded.

Work on several parts

Run branches in parallel and continue when every branch is done.

Send it back for another pass

A review can return the work for revision along the path you configured.

Change the routine for the next run.

Publish an improvement when you’re ready. Work already running keeps the version it started with; new runs use the update. Revoking access still takes effect immediately.

Refund request · versions

Run #104 keeps its starting versionv1
Next run uses the published updatev2

The process in the app

The refund process from the marketplace, imported into a workspace.
The return-and-refund process on the map: the agent's step, then tasks and an approval in one lane per role
The refund process on the map, one lane per role.
The process map beside the PROCESS.md file it is drawn from
The map and PROCESS.md are the same process. Edit either one.

Before you build

Can my agent help build the process?
Yes. Describe the outcome, inputs, owners and proof to your agent and it writes the PROCESS.md. You refine it as text or on the process map, which are two views of the same file. Publishing checks the file and lists everything it still needs before a run can start.
What if the work needs branches, loops or several people?
A process is one PROCESS.md file: steps for agents and people, labelled routes that the agent or person doing a step chooses with a reason, parallel branches that join, and approvals that can send work back to an earlier step. Tasks go to a person or a group, with due dates. The process map is a view of the same file, and you can import and export it.
What does an AI check actually do?
Configured AI checks assess outputs against your criteria and return a category, score or likelihood. An unclear or unavailable required check does not pass on its own; it goes to a person or back for rework. Agents cannot overwrite check results. Submitted evidence is not a guarantee that an agent’s claims are true.
How do testing and starting a run work?
Start a run in test mode first. Test runs are labelled, and they never send real emails or call your live systems. A person or an agent starts a run. The supplier onboarding process in the documentation is a complete worked example; the research flow above is an illustration you could write.

Current pilot limits. Scheduled and event-triggered starts are not switched on yet. Agent Process coordinates the process; your connected agent does the work.

Start with one routine you repeat

Join the waitlist with one recurring job in mind. When invited, we’ll help you build the process and try it with your agent. Work solo or with a team; the pilot is free.

Join the waitlist