Relay
Relay sits between IDE coding agents and the tools they want to call. Bash, path, MCP, and prompt controls evaluate before side effects—engineers keep velocity; the org keeps the audit trail.
Product hub: exemplar.dev/relay . Pricing: exemplar.dev/pricing .
Relay is Exemplar’s coding-agent governance module—not third-party Agent Relay. It pairs with Harness for production AI operations. Buy either alone or Full Platform for both.
Three jobs
- See — Inventory governed tools and active sessions across Cursor and Claude. Know what agents are allowed to reach before you need a forensic hunt.
- Control — Four channels evaluate every agent action before it runs: bash, path, MCP, and prompt. Denied calls return a clear reason; allowed calls keep speed.
- Prove — Tamper-evident events capture who ran which action under which policy, with what outcome.
How it works
Relay sits on the wire between your editors and your company. It doesn’t run actions itself—it reads each one, checks org policy, and returns a verdict.
| Step | What happens |
|---|---|
| An agent acts | Shell, file read/write, MCP tool, or prompt |
| The hook fires | Relay intercepts before side effects |
| Relay decides | Rules and policy packs return a verdict in milliseconds |
| Editor proceeds | Run, pause for confirm, or block |
| Prove | Every decision is recorded for audit / replay |
Topics
Related
- Harness — Gateway, prompts, skills, memory, and guardrails for production agents
- Account Settings — Tokens and team access
- Copilot for DevX Assist — MCP connection patterns in the IDE
Last updated on