SeenRelay gives AI agents a decision boundary before revalidating known external state. It checks local/native controls first, can add recent shared evidence, and falls back to the authoritative source. Free, no account, no API key.
I have built SeenRelay because I noticed that AI agents may repeat paid or slow read-only verifications.
I didn't want to simply build another 'cache', but a system that first tries to see if it is indeed useful and needed, before actually being implemented, so I made available the command: npx seenrelay scan .
It runs locally, doesn't upload code, doesn't modify your project and doesn't activate reuse.
It can output only three results: USE / DO NOT USE / INSUFFICIENT EVIDENCE.
USE — the measured workload shows that SeenRelay can add value after stronger local or provider-native controls have been considered.
DO NOT USE — a simpler or stronger native approach is better, so SeenRelay should stay out of that workload.
INSUFFICIENT EVIDENCE — there is not enough real workload data yet to justify either conclusion.
DO NOT USE will appear if ETag or cache or provider-native controls are better, so seenrelay will stand aside.
SeenRelay is integrated into Claude Code repository-hosted pending Anthropic review, so not yet approved.
Feed-back and real repo testing would be highly appreciated instead of upvotes.