ProductProveClaimed vs verified
Check the fix, not the claim.
An agent saying "fixed" is a claim. Someone still has to check that the change went in and the checks passed, and with dozens of agents that check gets skipped.
agent-3 Claimed
- … change merged
- … required checks
- … safe version in place
agent-5 Verified
- ✓ change merged
- ✓ required checks passed
- ✓ safe version in place
Design preview with simulated data. In development; not running yet.
How it works
Three steps, and your agents keep running where they already run.
The agent reports the fix
When an agent upgrades a risky library, it reports the change it made and links to it.
Hyoid checks the record
Hyoid looks up the facts in your code host: was the change merged, did your required checks pass, is the library now on a safe version.
Verified, or still a claim
If every check holds, the outcome turns Verified, with links to the evidence. If not, it stays Claimed and says what's missing.
What's included, and what's coming
Every item carries a label. We move a label only when the work ships. Hyoid Sessions isn't generally available yet. Early access: built and tested, opening first to people on the waitlist. Preview: built, still in testing. In development, Planned and Coming soon: not built yet.
Included
- At launch, Verified covers one kind of fix: upgrading a risky library to a safe versionIn development
- Checked against your code host's records (one host at launch)In development
- Claimed and Verified side by side on the session pageIn development
- No AI judging: the check is a lookup of recordsIn development
Coming
- More kinds of fixPlanned
- Evidence from CI, deploy and monitoring toolsPlanned
- Cost per verified fixPlanned
- Verified outcomes on share linksPlanned
Questions
Does an AI decide whether the fix worked?
No. Hyoid reads facts from the systems your team already relies on, such as whether a change was merged and whether its checks passed. No model gives an opinion.
What exactly does Verified mean at launch?
For a library upgrade: the change was merged, your required checks passed, and the library is now on a version that isn't flagged as risky. It's as strong as your own checks.
What if the checks fail?
The outcome stays Claimed, or is marked failed, with the reason shown. Nothing is counted as fixed on the agent's word alone.
Can the agent mark its own work as verified?
No. The connector that reports agent activity can only send claims. Verification comes from Hyoid reading the record.
When can I use it?
It's in development. Early-access teams will get it first.
See it on your own agents.
Join the waitlist by email and tell us which agent tools you run.