hoop.dev

# hoop.dev

> Data and context make AI useful, hoop.dev makes it safe. Hoop sits between engineers using coding assistants / agents and the resources the data, APIs, and services, masking sensitive data from the model and scoring each action for risk before changes are made.

hoop.dev is a proxy that sits between your agents and production. Every query and command gets a decision as it runs: allow, block, mask, or ask a human. The sidecar is free and open source under the MIT license. Enterprise adds a control plane that writes rules once and delivers them to every sidecar you run.

## Docs

- [Docs index](https://hoop.dev/docs/llms.txt): Every documentation page with a one-line summary. Start here for how hoop.dev behaves.
- [Docs, full text](https://hoop.dev/docs/llms-full.txt): The complete documentation in one file.
- [Getting started](https://hoop.dev/docs/introduction/getting-started): What the sidecar is and what it does to traffic.
- [Quickstart](https://hoop.dev/docs/introduction/quickstart): Run a sidecar and point an agent at it.
- [Architecture](https://hoop.dev/docs/introduction/architecture): The components of a deployment, how they connect, and how each tier scales.
- [Guardrail rules reference](https://hoop.dev/docs/setup/configuration/hoop-sidecar/policy-rules): Write a rule, pick deny or defer, hand the judgement to Rego.
- [AI analyzer reference](https://hoop.dev/docs/setup/configuration/hoop-sidecar/risk-analysis): Classify statements with a language model and block on the risk it reports.
- [Approval status for agents](https://hoop.dev/docs/setup/configuration/hoop-sidecar/agent-approvals): How an agent learns when a person decides a held statement.

## Product

- [How it works](https://hoop.dev/how-it-works): Agents connect to a different port. A proxy reads every query before it runs, masks sensitive fields in the response, and refuses destructive statements.
- [Data masking](https://hoop.dev/controls/data-masking): Sensitive values are rewritten in the response, in memory, before they reach the client. The request is never touched.
- [Guardrails](https://hoop.dev/controls/guardrails): An ordered deny list checked against every statement. Destructive ones never reach the database.
- [Human in the loop](https://hoop.dev/controls/human-in-the-loop): Risky operations pause for a person's approval. The agent waits with the exact statement attached, then continues.
- [AI analyzer](https://hoop.dev/controls/session-analyzer): An agent scores every action for intent and syntax before it executes. Not smarter than the agent it watches, just as fast, with one job.
- [Claude Code](https://hoop.dev/claudecode): What Claude Code can see and do, decided at the protocol layer.
- [Get started](https://hoop.dev/start): Install the sidecar. One binary, one config file, free.
- [Pricing](https://hoop.dev/pricing): The open-source sidecar and what Enterprise adds.

## Source

- [hoophq/hoop on GitHub](https://github.com/hoophq/hoop): The open-source repository.

## Company

- [About](https://hoop.dev/about): Who builds hoop.dev.
- [Manifesto](https://hoop.dev/manifesto): Why we govern the wire instead of cataloging the world.
- [Blog](https://hoop.dev/blog): Engineering deep-dives, security research, and product thinking.
- [hoop University](https://hoop.dev/learn): Short lessons on operating hoop.dev.
- [hoop.labs](https://hoop.dev/labs): Free tools for teams running AI agents: guardrails, credential isolation, PII detection, session risk scans, ROI modeling.

## Optional

- [White papers](https://hoop.dev/start/whitepapers): Long-form technical papers.
- [Webinar: How do you know AI did what you intended?](https://hoop.dev/webinar/ai-accountability): Recorded session on AI accountability and runtime governance.
- [Talk to the team](https://hoop.dev/meet): A 30-minute session with the engineers who built hoop.dev, with a demo on your stack.