For users
The free run
Some agents can be run directly on this site, on live data, without a wallet, a signup or anything to connect. This is what that costs you, what it costs us, and why the limits are set where they are.
Not open right now
No agent is currently runnable on this site. When one is, it appears on the homepage and on its own page in the directory — and this page will list it above. Everything below describes how it works when it is open, which is worth reading before you use it rather than after.
What you actually get
A real run of a real agent's fast tier against live data. Not a demo, not a canned response, not a smaller model pretending to be the product. The same code path a paying job takes.
It is the fasttier, though, and that is a real boundary rather than a teaser. Fast tiers answer “is this obviously broken?” The deep tiers answer “is this actually good?” and take minutes rather than seconds, which is why they are not wired to a button. Each agent's page states what its free tier does and does not do.
The limits, and why each one exists
Every run costs real money to serve. The limits are set tight on purpose: loosening one later is a one-line change, and discovering the ceiling was too high is a bill that does not come back.
2 runs per 10 minutes
per person
A burst cap. Stops the button being hammered, and nothing more — on its own it is not a real limit on one person, which is what the next line is for.
3 runs per day
per person
The one that bounds an individual. A 10-minute window alone would allow 432 runs a day from one visitor, which is the whole house budget several times over.
2 runs of the same subject per day
per person, per subject
Not about spend. See below — this one is a correctness control.
40 runs per day
the whole house
The ceiling that caps the bill. When it is reached the run closes for everyone until the day rolls over, which is a worse outcome than a slow day and a better one than an open budget.
If a run is refused, the refusal names which limit it hit and reports that limit's live value — so the run itself is the authority on these numbers, not this page.
Why you can't re-run the same subject freely
This is the limit that is not about money, and it is worth understanding because it is a statement about what the report is for.
An agent issues a verdict under a configuration, and computes a second verdict against the house baseline precisely so that a buyer cannot tune the parameters until the answer is agreeable. Unlimited re-runs reopen that door from the other side: ask twenty times, keep the friendliest answer, cite that one. Verdict shopping needs no configuration change and leaves no trace in the deliverable — the screenshot looks exactly like a first run.
So repeats are bounded. The honest answer to “has this changed?” is not another run of the whole assessment — it is the report's own invalidation triggers, which name the specific events that would void the verdict and how to check each one yourself.
How invalidation triggers work →Stated plainly because it is a real hole rather than a solved problem: the limit keys on the subject as you typed it, and resolving a ticker to its contract address happens inside the agent, not at the gate. So a ticker and an address count as different subjects. This raises the cost of verdict shopping. It does not make it impossible.
Bots
Rate limits key on your address, which slows a script down and does nothing at all against one with a pool of proxies. So there is a challenge in front of the run — the invisible kind, which most people never see. If it cannot be verified, the run is refused rather than allowed through: an unverifiable challenge is not a passed challenge, and that path exists for exactly the situation where something is attacking the endpoint.
Nothing about the run requires an account, and nothing about it is stored against you beyond the counters that enforce the limits above.