Lab.
The private console where bounded campaigns are planned, run, inspected, and approved.
Lab has three areas: Lab, Results, and Settings. A campaign starts with a frozen evaluator and a budget, can be paused, cancelled, or stopped at once, and charges every call to a cost ledger capped at $50 a month. Controls and logs stay private.
Runs today, privately. Nothing has been publicly approved yet.
See an example
A campaign is more than its result.
Lab keeps the question, the evaluator, the budget, the cost, and every attempt together, so an exact version can be reviewed before anything is published.
How do you let an AI run experiments without losing control of the cost, or of what gets published?
The problem
An agent that does research can spend money, quietly change its own test, or report a lucky result. Without bounds, you can trust neither the bill nor the claim.
Why it is hard
Useful campaigns are long and mostly fail. At any moment someone must know what was frozen, what it has cost, and which exact version would be made public.
How Mutome approaches it
Lab fixes the rules before the first call, charges every call to a capped ledger, keeps every attempt, and publishes nothing until a person approves one exact, hash-bound version.
The life of one campaign
Follow one bounded campaign, here a synthetic oscillator task, from a draft to the publication decision.
What runs today
Bounded campaigns with frozen evaluators and budgets, leases, pause, cancel and global stop, the cost ledger, replay, artifact hashing, backup and restore, and hash-bound approval and retraction. Campaigns so far were synthetic; most attempts were discarded or inconclusive.
Not yet
Lab does not yet run on its own around the clock, and no campaign has been approved for publication. The next step is to run Slean labs continuously within the same $50 monthly budget.