Sentinel v0.1.0 Public Preview

vedantggwp released this 01 Jun 01:34

· 21 commits to main since this release

8b9a7e8

Sentinel v0.1.0 Public Preview

Initial public preview of Sentinel, a safety and claim-verification layer for
sponsored recommendations inside AI conversations.

Included

FastAPI /v1/analyze endpoint for ad placement verification.
MCP verify tool using the same safety pipeline.
Deterministic APPROVE, BLOCK, and ESCALATE gate.
Signed receipt support when an ed25519 key is configured.
Local audit trail and trace-console demo assets.
Seed eval regression and adversarial held-out measurement.
Security policy, contributing guide, maintainer file, roadmap, and GitHub CI.

Verification

GitHub Actions CI: https://github.com/vedantggwp/sentinel/actions/runs/26730578617
python -m pytest -q passes 92 tests in CI.
python -m sentinel.eval reports seed regression 25/25.
Adversarial held-out set is reported as measurement only and currently passes 3/10.

Known Limits

Tavily live claim verification, Overmind span export, and Thrad bid
normalization are public-v1 roadmap items with fixture or optional paths today.
No broad adoption is claimed for this release.
External services must never bypass the deterministic placement gate.

Assets 2