- Customer workflows mapped to agent behavior
- Up to five implementation-ready changes
- Private comparison against up to three competitors
- Official retest after implementation
WHAT DISCRY DOES
Make your API findable—and understandable once found.
Discry connects behavioral evidence to a tested change. A file such as llms.txt matters only when it helps an agent find the right documentation and complete a real task.
THE PRODUCT LADDER
Measure the failure. Ship the fix. Protect the next release.
Start with the public behavior of your API. Move from evidence to implementation only when the scan finds a problem worth fixing.
- 01
Free Scan
Free
Discovery probes and graded comprehension tasks in one run: which discovery files agents look for and cannot find, where your docs misled them, and where you stand in the current cohort.
Scan your API → - 02
Fix Kit
$119 one time
Turn failed behavioral checks into specific, tested changes your team or coding agent can implement.
Start with a scan → - 03In development
Discry on Your Repo
A GitHub app that checks every docs pull request against the agent tasks it affects and opens a fix as a PR.
Join the waitlist →
FIX KIT
A repair package for the failures agents actually hit.
You get specific changes, exact implementation instructions, a portable manifest, disclosed pass evidence, and a kickoff prompt for the coding agent of your choice. Current-cohort context is included when the comparison is valid, so your team can prioritize the work.
DISCRY ON YOUR REPO
Catch agent-readiness regressions before docs ship.
- Runs the discovery probes on every docs pull request, so a deleted llms.txt or a moved AGENTS.md never reaches production.
- Re-runs the comprehension tasks the changed pages affect, so you see which agent tasks regress before the docs ship.
- Opens a pull request with the proposed fix. It never edits your docs on its own; your team reviews and merges.
- Re-scores the API after each notable model release, so a model change that breaks comprehension shows up in your repo, not in a customer’s agent.
Richer implementation patches and comparisons across model changes come later; neither is available today.