hoop.dev

hoop.dev offers runtime controls for AI agents in the form of a lightweight sidecar proxy that attaches to an existing resource — for example a PostgreSQL database, a Microsoft SQL Server instance, or an agent sandbox — and sits on the wire to inspect queries, commands and responses as they occur. The company positions the product as requiring no new infrastructure, no code changes and no rerouting of traffic, reusing the customer's existing credentials while leaving privileged access management, single sign-on and identity provider systems in place.

The product's stated capabilities fall into four areas: data masking, which redacts regulated or sensitive fields in flight across protocols before an agent sees them; deterministic guardrails that automatically block dangerous operations; a session analyzer in which an agent scores each action's intent and syntax for risk prior to execution; and reviews, which escalate risky operations to a human approver for one-off approval via Slack, Teams or Jira. The company states each decision is made in under 5ms, before the request reaches the resource.

Marketing material frames the use cases around unblocking AI projects held up by compliance — for instance allowing an agent to query a transaction database while personally identifiable information is masked at the wire and an audit trail is produced — and around replacing per-source MCP server sprawl with a single sidecar that covers every protocol, including legacy protocols not supported by MCP.