Build a Hoop guardrail regex from a query you want blocked, test it live against your own examples, and export it ready to enforce. Everything runs in this page; your queries never leave the browser.
The CLI proves rules in CI (warden test) and drafts them with an LLM (warden generate).
Paste a query you want to block, plus an optional similar one that must stay allowed. Warden recognizes the shapes and suggests vetted rules, pre-filled from the queries themselves.
Pattern rules match the text of a query and cannot parse SQL. A rule that passes each test here can still be evaded by a rewritten query (CTEs, subqueries, obfuscated conditions). Pair guardrails with review for high-risk access.
Matching runs in your browser with RE2, the portable subset both of Hoop's guardrail engines accept. It proves behavior on these examples only; untested queries are untested.
Build a rule first; the exports appear here.
How it works
Paste a query you want blocked, pick a vetted template, or write raw RE2. Or let an LLM draft it on the command line.
The test bench runs every example through the same RE2 engine Hoop compiles. In CI, the CLI does the same and fails the build on any regression.
Export a ready-to-run curl that creates the guardrail through Hoop's API, then attach it to your connections.
One rule blocks one query shape. hoop enforces the whole policy.
Guardrails, data masking, access control, and full audit logs, enforced centrally across every agent, every connection, and every protocol. That is the hoop.dev platform.