Emergency Stops and Undo Buttons
for your Agentic Workflows
Relay sits between your agents and their tools to prevent mistakes, stop them mid-execution, and undo the damage.
Production agents are hard to control.
When an AI agent makes an authorized mistake in production, there's no way to:
- ×Stop it mid-execution
- ×Pause risky actions
- ×Undo what already happened
- ×See what else it affected
Logs tell you what happened.
Permissions tell you what's allowed.
Neither protects you in real time.
Built for teams deploying AI agents
Relay is shaped by what production teams actually need: safety, control, and confidence.
When agents touch real systems, logs and authorization aren't enough. Teams need real-time protection and recovery.
Detect mistakes before they happen
Stop disasters instantly
Undo damage when it slips through
Fits into your existing workflows.
Relay integrates without rewriting your agents and gives you full control over what they can do.
Frictionless integration
Relay works with LangChain, CrewAI, Azure AI Foundry, and any other AI framework.
Relay SDK
Wrap existing tools. No agent logic changes. You code. Relay runs beside you.
Execution proxy dashboard
Live agent activity, stop agents instantly, review risky actions.
Human-in-the-loop controls
Approval queues, edit before execution, deny or stop agent.
Guardrails and policies
Rate limits, value thresholds, protected resources, time-based rules.
YAML-based configuration
Define safety in config, not code. Version-controlled policies.