Documentation
¶
Overview ¶
Command seedgen writes the performance seed dataset: N published agents with realistic names, summaries, manifests, capabilities and install counts, straight into the database through the store.
The search baseline is "p95 < 300 ms at a 10k-agent seed dataset". A baseline is only a baseline if the dataset it was measured against can be reproduced, so this generator is deterministic: the same -count and -seed always produce the same corpus, including the install-count distribution that drives search ranking.
It is staging/dev tooling. It is not run by CI, it writes no archives, and it never invokes the scan pipeline — versions are inserted already published, because the point is to populate the read paths (search, resolve, listing), not to exercise the gate.
go run ./scripts/seed -dsn "$AGENTRY_DATABASE_URL" -count 10000