hoop.dev is an open-source Layer 7 gateway that its makers position between an AI coding agent and the systems the agent acts on. Rather than letting the agent talk directly to a Git server, container registry or CI/CD orchestrator, the gateway receives the request, inspects the payload at the protocol level and applies policy before forwarding anything downstream — and where the policy calls for it, blocks the forward flow while an authorized reviewer decides.
The vendor’s own argument for this placement is that identity alone is not enforcement. When an agent authenticates directly against a target, the setup verifies who is calling but leaves no point at which a policy can pause, inspect or require a human decision, so the operation is recorded only in the target’s logs, if at all. The vendor names hallucinated secrets committed to source files, dependency updates carrying known CVEs, and logic that bypasses existing security controls as the kinds of thing that reach version control under that arrangement, and argues that a post-hoc pull-request review is too late because the artifact already exists — potentially exposing secrets or triggering downstream jobs before a reviewer intervenes.
Capabilities
The gateway’s stated behaviour for a Git push is to intercept it, extract the diff, and check the configured policy. Where that policy includes Human-in-the-loop approval, the diff is routed to a reviewer’s console and the commit is forwarded only after explicit consent. The session is recorded in full and the approval decision stored alongside it.
Because the gateway is the only component that sees the request, the vendor describes several controls as available only from that position: inline masking of detected secrets before they reach the repository, rewriting insecure imports, blocking, and logging every interaction for later replay. It also holds the credential for the target service itself, so the agent never sees the underlying secret — which the vendor presents as a reduction in credential-leakage risk rather than as an access-control change.
Policy is driven by identity. The described setup begins at an identity provider such as Okta or
Azure AD; the agent authenticates over OIDC, and the gateway verifies the token and extracts group
membership and attributes. The worked example restricts requests to agents in an ai-coding group
and consent to reviewers in a code-approval group.
Adoption & Ecosystem
hoop.dev is MIT licensed and available on GitHub, so it can be self-hosted inside an organization’s own network and wired to an existing identity provider. The vendor states it works at the protocol level, so any agent that can speak Git, HTTP or the relevant API can be routed through it, and that it sits in front of existing CI/CD tooling rather than replacing it — adding approval and audit without changing the downstream toolchain.
The audit trail is the outcome the vendor puts forward for compliance: session logs, a record of who approved each change, and the stored masked diff, exportable to meet reporting requirements.