Skip to main content
An experiment is a launched comparison. You select tasks, agents, and treatments. Oqoqo runs each combination and shows the results. You can define and launch from the dashboard, Ask Oqo, the CLI, or MCP. Results appear in the same place. Files and a machine attach to each task. Skills, MCP servers, CLIs, and SDKs attach to treatments. Agents are chosen at launch. See Core concepts.

​
Start points

  • New experiment for a blank draft
  • A Home template for a ready-made draft
  • Clone experiment on a finished experiment to reuse its plan

​
What you select at launch

Each combination is one billable run.

​
Workflow

1

Define the work

Write a task and a rubric. Attach files. Pin a machine, or keep the Standard Dev Machine.
2

Define the conditions

Create assets under Manage → Assets. Attach them to treatments. Keep a baseline treatment, usually the raw agent, so you can measure lift.
3

Launch

Select tasks, agents, and treatments. Set trials and Max Concurrency. Turn on Evals and Insights if you need rubric scores and frictions.
4

Compare

Every run of the same task starts from the same files and machine. You measure treatment, agent, and model. Read Matrix, Frictions, and run detail.
While an experiment is running, you can Cancel experiment to stop unfinished runs. Completed runs stay available. See Launch options.

Templates

Start from a ready-made draft.

Tasks and rubrics

Write the work and how you score it.

Launch options

Trials, concurrency, clone, and cancel.

Reading results

Matrix, Runs, and Frictions.