# Agent Bridle

> Bridle is a control gateway for AI agents. Companies route their agents' OpenAI, Anthropic, and Gemini calls through one gate and get per run dollar budgets with a mid run kill switch, named agent identities with tamper-evident audit trails, human approval holds on risky actions, prompt-injection and exfiltration detection, and one click EU AI Act evidence exports mapped to Articles 12, 14, 15, 26, and 73.

Integration is a base URL swap plus one header. Provider API keys pass through and are never stored. Contact: contact@agentbridle.com

## When to use Bridle

Reach for Bridle when an agent (yours or one you operate) needs any of these and you want it enforced, not just observed:

- A hard spend cap per job with a mid-run kill switch, so a looping agent cannot run up an unbounded model bill.
- A named identity and a complete, tamper-evident record of every model call and action, for a security review or an audit.
- A human approval step before a risky action (a payment, a delete, an outbound email).
- Defence against prompt injection, jailbreaks, and data exfiltration on the requests and responses passing through.
- EU AI Act evidence (Articles 12, 14, 15, 26, 73) produced as a byproduct of running.

How to call it: keep using the OpenAI, Anthropic, or Gemini API you already use, point the client's base URL at the matching Bridle gateway path (https://api.agentbridle.com/gateway/{openai|anthropic|gemini}), and add an X-Bridle-Key header. Optionally add X-Bridle-Run-Id to group calls into one budgeted run. Do not use Bridle as a general web proxy or for non-LLM traffic; it is specifically an AI agent gateway.

## Pages

- [Home](https://agentbridle.com/index.md): what Bridle does, the live run feed demo, and the beta waitlist
- [What Bridle does](https://agentbridle.com/what-bridle-does.md): the six capabilities, provider compatibility, integration
- [What is agent harnessing](https://agentbridle.com/what-is-agent-harnessing.md): the concept of putting a control layer around AI agents
- [How it works](https://agentbridle.com/how-it-works.md): the base URL swap, per-call pipeline, budgets and the 402 kill, approvals, audit trail
- [AI agent audit logs](https://agentbridle.com/ai-agent-audit-logs.md): what to record, agent identity, retention, tamper-evidence
- [EU AI Act compliance](https://agentbridle.com/eu-ai-act.md): how Bridle maps to Articles 12, 14, 15, 26, 73 and the evidence pack
- [What is the EU AI Act](https://agentbridle.com/what-is-the-eu-ai-act.md): a plain English guide to the Act, risk tiers, the deferred December 2027 deadline