Skip to main content
Oqoqo turns an agent workflow into a repeatable, measurable experiment. This guide walks you through every step — from requesting access to reviewing your first results. The whole process takes a few minutes if you use a template, or slightly longer if you’re defining a custom task from scratch.
Start with a template task from the Oqoqo library to skip setup and run your first experiment in under 5 minutes.
1

Join the waitlist

Go to oqoqo.ai/waitlist and request access. You receive up to 100 free runs — enough to explore the platform thoroughly before committing to a plan.
2

Open the Lab

Once you have access, log in to the Oqoqo dashboard and navigate to Lab › Define Experiment. This is where you configure every aspect of an experiment before launching it.
3

Define your task

Write a natural-language instruction that tells the agent what to accomplish. Be specific about the goal, the expected output format, and any constraints.Example task:
Open five product pages in parallel browser tabs, extract the price from
each, and return them as a JSON array in page order.
Optionally, click + Add rubric to define evaluation criteria — for example, “Returns the correct shape,” “Preserves page order,” and “Reports cache state.” Oqoqo automatically rewrites your rubric to follow best practices before it’s applied.
4

Select agents and treatments

Choose which agents run your experiment and at what settings. You can combine multiple agents and effort levels in a single experiment — for example:
  • Claude Code Opus · High
  • Claude Code Sonnet · High
Then choose the treatments you want to compare:
  • Baseline — the agent runs with no additional context or tooling
  • + skill — the agent receives a skill that describes your interface
  • + MCP — the agent connects via your MCP server
Oqoqo runs each agent–treatment combination as a separate track within the experiment, so you can compare results side by side.
5

Configure your library and environment

Attach the repos and data fixtures the agent needs under Library. These are checked out into each sandbox before the agent starts, ensuring every trial starts from an identical state.Under Environment, choose where sandboxes run:
  • Local — runs on your own infrastructure
  • Cloud — Oqoqo provisions and manages the environment for you
6

Set trial count and launch

Set the number of parallel trials (1–N) you want to run. More trials give you statistical confidence in your results; fewer trials are faster and cheaper for early exploration.Click Launch. Oqoqo spins up a clean, isolated sandbox for each trial and starts running immediately.
7

Review results

When trials complete, open the experiment to review:
  • Traces tab — the full step-by-step trajectory for each run: tool calls, commands, token counts, errors, and friction points
  • Output tab — the final answer or artifact produced by each trial, plus a diff of any files changed
  • Evals tab — pass/fail outcomes scored against your rubric, with per-criterion breakdowns
Compare metrics across runs — steps, tokens, cost, duration, tool calls — to understand which agent, model, or treatment performed best on your task.
Unfamiliar with terms like trials, treatments, rubrics, or traces? See Core Concepts for a complete glossary of Oqoqo building blocks.