ComplyEdge

Compliance tools for AI agents usually live outside the request path — they scan logs after the fact, produce confidence scores, and leave you explaining to a regulator why a number like 0.87 constitutes a policy decision. ComplyEdge sits inside the request path instead. The Python SDK exposes a `@compliance_check` decorator that wraps agent functions; every call is evaluated against a rule engine, and blocks return a structured response citing the specific Article 5 provision, a rule ID, an input hash, and a timestamp. The audit trail is the output, not an afterthought.

The differentiating design choice is explicit citation over probabilistic scoring. Where most compliance layers return a risk score, ComplyEdge returns `Article 5(1)(a), rule rego-art5-1a-001` — the kind of reference a legal team can put in a regulatory filing. The vendor positions this directly against the GPAI fine timeline (August 2, 2026 per the repo), framing the audit trail as the compliance artifact rather than a supporting data point.

TrustLint, shipped as a separate pip package under `packages/trustlint`, brings the same rule evaluation into offline development pipelines. Teams can run compliance scans in CI before any code reaches a production agent, catching Article 5 violations at the diff stage rather than in a live request. The self-hosted, Apache-2.0-licensed architecture means no data leaves your infrastructure — a requirement for many regulated environments.

The documented scope is EU jurisdiction and Article 5 enforcement. The repo structure shows `rules/`, `providers/`, and `sdks/python/` directories, but the scraped page does not confirm pre-built rule sets for other regulatory frameworks. Teams with obligations beyond EU AI Act Article 5 — GDPR data minimization, sector-specific financial rules, US state AI laws — will need to author and maintain their own rule files. Non-Python agent stacks have no documented SDK path at time of publication.