Skip to main content
At launch you choose which tasks, agents, and treatments to cross. Then you set repeats, concurrency, and whether to run Evals and Insights. The same options are available in the dashboard, CLI, and MCP.

​
What you are crossing

You do not attach files, machines, or assets at launch. Those already live on the task or the treatment. See Core concepts and Supported agents. Start faster from a Home template when you want a ready-made draft.

​
Trials

Trials are repeats of the same inputs. Two is a good start. Each repeat is one run from your organization’s balance.
See Reading results and Billing.

​
Max Concurrency

Max Concurrency limits how many agent sessions can be active at once. Each session is one trial of one task × one agent × one treatment. Use a lower limit when sessions share tools that must not race. Set 1 to run sessions one after another. This setting does not control Evals or Insights. The effective limit can also be lower because of organization and global caps.

​
Evals and Insights

Assign the models under Oqoqo AI features. Turn either on independently, or trigger later. At launch you can confirm or override the models used for Run evals and Insights for that experiment. See Evals and Insights.

​
Clone an experiment

On a finished experiment, choose Clone experiment. Clone opens a new experiment draft with that experiment’s tasks, treatments, and settings. Review or change anything before you launch. If newer task versions exist, Oqoqo asks you to review versions before cloning. Treatments use the current saved state. Clone is available when the experiment is completed, completed with errors, errored, failed, published, or cancelled. From the CLI or MCP, use experiments clone-plan / get_experiment_plan to reconstruct a launchable plan. That does not launch automatically.

​
Cancel an experiment

While an experiment is running, choose Cancel experiment. Cancel stops unfinished runs. Completed runs stay available. You can also cancel from the experiments list menu, or with CLI/MCP experiments cancel.

​
After launch

  • Matrix — pass rate and lift
  • Frictions — where agents got stuck
  • Runs — every execution
Use Ask Oqo for a plain-language read of the same recorded work.