Build with Carver

Yes, see the GitHub and Documentation links above for the Python SDK, which includes a Query Engine for pulling regulatory signals and an Annotations module for structuring/labeling data, so you don't have to hand-roll HTTP calls against the raw API.

New regulatory content typically shows up within 24 hours of publication by the source institution, so API responses reflect near-real-time regulatory activity, not a stale batch feed.

You can pull Signals (scored regulatory changes), Statutes (the underlying laws/guidance library), Risk Intelligence Reports (AI-generated PDF outlooks), and — for agent builders — agent-specific evals and RegTeaming results, all linked back to their source statute so you can trace any output to the underlying regulation.

RCG exposes regulatory context as a graph specifically for AI agents — at dev time (grounding agent design and decisions in current regulation) and at run time (agents checking live regulatory context before acting). It's built for agentic systems that need to stay compliance-aware continuously, not for static dashboards on a refresh schedule.

Summaries are generated from original source text, and relevance scores (0–10) depend on the Context you configure (your Business Profile + Regulatory Focus Areas) — better-defined context means sharper scoring. Carver recommends verifying critical regulatory details against the original source before acting on them.

Yes, check out our demo playlist covering authentication setup, sample Signals/Statutes queries, Python SDK usage, and the RegTeaming method end-to-end. It's the fastest way to see the API and SDK working in a real integration before you start writing code.

RegTeaming is an adversarial testing methodology, sponsored and maintained by Carver, that checks whether an AI system can be pushed into unlawful, prohibited, or regulatorily indefensible behavior. Red teams test whether a system can be made unsafe; RegTeams test whether it can be made non-compliant. It's not a replacement for safety red-teaming — it's the regulatory layer most eval stacks don't cover yet, since regulatory failures emerge across a full trajectory: what the agent says, which tools it uses, which disclosures it makes, when it escalates, what action it ultimately takes, and what evidence gets preserved.

RegTeaming converts changing regulatory obligations — sourced from the same regulatory graph behind RegWatch and RCG into adversarial scenarios, expected behavior, evidence requirements, and repeatable evals. A rule change becomes a concrete eval against your agent's behavior instead of a line in a policy document. The methodology follows an eight-principle manifesto (RegTeaming Manifesto v1): test behavior not declarations, test control effectiveness, preserve evidence, update continuously, report scope honestly, among others.

Evals are generated directly from the underlying legal instruments — bills, directives, acts, and similar sources — by breaking them down into their component clauses, obligations, and other structural elements. Each of those elements can then be turned into a concrete scenario an agent is tested against, rather than the eval being hand-written from a summary of the law.

Yes, RegTeaming is built to be used and extended by teams building AI assurance, legal, or risk tooling. The manifesto and developer guide are published at regteaming.com, along with the scenario/evidence framework you can adapt to your own agent trajectories and compliance obligations.