Hoop.dev describes itself as a Layer-7 access gateway for humans and agents. AI agents inherit the human user's identity, scopes, and audit trail when they connect to production data sources, so the same governance system covers both. The implicit category claim is that the constraint on enterprise AI-agent deployment in 2026 is not model capability or tool-calling fidelity. It is whether the security organization will let an agent touch a Postgres replica without a wrapper of audit logs, masking, and revocable scopes around the connection. Hoop builds that wrapper at the wire.
The product is a transparent proxy that sits between an agent (or a human engineer) and the underlying data source. It speaks PostgreSQL, MySQL, Kubernetes, SSH, gRPC, and a list of other protocols natively, so existing tooling does not need to know it is there. Three primitives carry the value proposition. Inline data masking redacts PII, credentials, and tagged sensitive fields before any response reaches the agent, meaning the model never sees the raw record. Guardrails block destructive commands at the gateway, letting an agent continue working on permitted actions instead of having the connection torn down. Command-approval workflows route high-risk requests to Slack, Jira, or Microsoft Teams for a human sign-off before they execute. The result is closer to a least-privilege session manager than to a traditional database firewall.
The company is headquartered in Boston, with engineering distributed across the US and Brazil. Hoop closed a Seed round on October 29, 2025, led by Venture Guides and backed by Y Combinator, the most recently disclosed financing event and the round that puts the access-control thesis on a venture-priced track. Early design partners include PicPay, Unico, EBANX, RD Station, and Dock, fintech and SaaS operators running heavy compliance workloads that pushed the gateway to harden against the audit, residency, and least-privilege requirements US enterprise security teams will impose next. Lucas Teske, Sandro Mello, Andrios Robert, and Michael Daniels show up in public team disclosures. Public sources put headcount in the 11 to 50 band. That sizing positions Hoop ahead of the curve, betting the access-control category hardens into a real venture market as agentic deployments accelerate inside US-based enterprises.
The bet sits inside a broader infrastructure thesis that frontier-tech investors are now writing checks against. AI safety capital has flowed to model evaluation, to red-teaming, to prompt-injection defense, and to interpretability research. Production agent access is the operational equivalent of those research bets, the layer where the abstract problem of agent autonomy meets a security team's specific requirement for revocable scopes, immutable audit trails, and pre-approved action sets. The companies whose AI roadmaps are not blocked by access risk are the ones who have already solved this with internal tooling. Hoop's pitch is that buying the gateway is cheaper than building it, and that the wire-level approach is the one that survives multiple agent runtimes, multiple model providers, and the inevitable agent-orchestration framework migrations of the next three years.
Three things to watch over the next twelve months. First, whether Hoop closes a priced Series A, because the round size and lead will reveal how the venture market is pricing the agent-governance category. Second, the pace at which US enterprise fintech and SaaS adoption accelerates, the proof point that turns Hoop's Boston-based bet into a category-defining piece of US enterprise infrastructure. Third, whether competing approaches, MCP server gateways, agent-runtime sidecars, or vendor-native solutions from Snowflake and Databricks, converge on or diverge from the wire-level model. The right answer to that question determines whether Hoop's gateway is a category or a feature.