RELAY

THE CONTROL PLANE FOR AGENTIC SYSTEMS

Control,containment,andrecoveryforAIagentsinoperation.

Relay sits between agentic systems and the tools they act on, interdicting unintended actions, halting execution in real time, and recovering from effects already in motion.

Request Access Request a Demo

THE GAP

CAPABILITIES01 / 03

Detection

Every action evaluated against policy-defined thresholds before it reaches a live system. Value limits, protected resources, and rate constraints checked in-line, in milliseconds.

CAPABILITIES02 / 03

Interdiction

Halt agents mid-execution. Pause, reroute to human review, or terminate across single agents, swarms, and downstream tool chains.

CAPABILITIES03 / 03

Recovery

Reverse effects after the fact. Automated rollback paths, compensating actions, and full audit trails of what executed, what was reversed, and what remained.

01 / 03 · Detection

PLATFORM

Vendor-neutral. Framework-agnostic. Deployable where the mission requires.

Relay integrates with the frameworks teams already use and runs in the environments those teams already operate in.

FRAMEWORKS

Frictionless integration

LangChain, LangGraph, CrewAI, Azure AI Foundry, Amazon Bedrock, and custom agent runtimes. Wrap existing tools without rewriting agent logic.

Relay SDK

Drop-in integration. Your agents, your tools, your code: instrumented for control without architectural change.

main.py

from relay import Relay

relay = Relay(proxy="relay.dev")

tools = relay.wrap({ send_email, refund })

Execution control dashboard

Live agent activity, real-time halt, action-level review, and intervention history.

Human-on-the-loop controls

Approval queues, edit-before-execute, denial, and termination. Configurable per agent, per tool, per action class.

Policy and guardrails

Rate limits, value thresholds, protected resources, time-window rules, and per-action authorization.

Configuration as code

Safety expressed in configuration, not application logic. Policies travel with the codebase, change under review, and produce a record.

The control plane for AI agents in operation.

Request Access Request a Demo

POLICY ENGINE

Safety expressed in configuration.

Relay policies travel with the codebase, change under review, and produce a full audit record. Define once, enforce everywhere.

relay.yaml

policies:
  - tool: send_email
    require_approval: true
    protected_domains:
      - "*.mil"
      - "*.gov"
    max_recipients: 10
    on_violation: halt

Deploy agents you can stop, contain, and recover.

Relay is in limited release with defense-tech integrators and enterprise platform teams. Request access or schedule a technical review.

Request Access Contact