Workflows

A workflow splits a large task into many small pieces and gives each piece its own agent. The agent in your session writes a short script that says which agents to start and what to ask each one. OpenChamber runs the script in the background, so your session stays free while the agents work. When the run ends, its report comes back into the session.

Workflows fit work with many similar, independent pieces: a review of every changed file, an audit of every API route, a migration across many files, or research where each source is checked on its own. A single feature or a few steps is faster in one session.

Turn it on

The workflow tool is off by default. Open Settings → General → OpenChamber Tools and turn on Workflow tool. Running sessions get it on their next turn. Turning it off again removes the tool and the instructions about it from every session and hides the Workflows panel; runs you already have come back when you turn it on.

Workflows need OpenChamber’s own OpenCode server. They are not available when OpenChamber connects to an external OpenCode server through OPENCODE_HOST or skip-start, or inside the VS Code extension.

Run a workflow

Ask for one in plain language, for example “Run a workflow that reviews every file changed on this branch and lists the bugs it finds.”
The agent proposes a plan and a model for each phase. Agree on the models before the run starts: see Models and cost.
A card with the plan appears in the chat. Select Run workflow to start it, or Don’t run to drop it. Nothing runs until you choose.
While it runs, the card shows each phase and agent. The session’s row in the sidebar says Workflow running, and the Workflows section of the work-status panel lists the run. Select Stop to end it early.
When the run ends, its report arrives in the session as a message. A finished or failed run wakes the agent so it can act on the result; a run you stopped or declined does not.

A stopped or failed run can be continued with Resume. Agents that already answered keep their answers, and only the rest run again.

Agents of a running workflow cannot start a workflow of their own.

Models and cost

Every agent in a workflow runs on your session’s model unless the script names another one. Dozens of agents on a top model cost many times what the same work costs on a small one. The agent therefore suggests a model per phase: a cheap one for reading and checking against a fixed list, a stronger one where a phase weighs judgement or writes what lands in your project. Change the choice before you approve the run if it does not suit you.

At most 8 agents work at the same time, and one run starts at most 200 agents across all its attempts.

The Workflows panel

The Workflows panel in the right rail shows the runs of the project: its agents, the log, the result and the script, with the model and token use of each agent. Open any agent’s session from there or from the chat card to read its full conversation. The icon appears once the project has a run or a saved workflow.

Workflow agents do not appear in the session sidebar, since one run can start a hundred of them. They stay reachable from the run’s card and the Workflows panel.

Delete run removes a finished run from the panel. The agents’ sessions stay.

Saved workflows

Ask the agent to save a workflow you want to run again. It is stored as .openchamber/workflows/<name>.js in the project, so you can commit it with the rest of the repository.

Run a saved workflow from Saved workflows in the panel’s run picker. It starts right away, in the open session or in a new one, and its report comes back to that session. You can also ask the agent to run it by name.