How the agent gateway works

The agent gateway puts agent-to-agent (A2A) traffic behind CoreValue. Your organisation registers the agents it uses. Callers reach each one at a CoreValue address with a CoreValue API key, and CoreValue forwards the call to the agent, with the credential you stored for it when the agent needs one.

A registered agent has:

• an agent ID, which is its path at the gateway;
• the agent's own URL, where CoreValue reads its agent card;
• the A2A version it speaks, 0.3 or 1.0;
• who may call it;
• optionally, a credential CoreValue sends to the agent;
• the CoreValue keys the agent itself calls with, so that its own LLM, MCP and A2A traffic is attributed to it.

An agent that is never called by another agent can be registered with no endpoint: only its ID, name and keys. See Register an agent.

What CoreValue records

Each call records the caller and the agent called, the method, the task and its state, the decision (allowed, refused, or served but recorded as one an audit rule would refuse), the timing and the size. A refused call is recorded too, so you can tell a refusal from an agent failure.

A2A calls are not priced yet, so the call itself carries no cost. The LLM and MCP spend an agent makes while it works on a task is recorded on its own rows, and is tied back to the task when the agent passes the task trace on. See Call an agent.

Access

Any member can view the agent registry and the agent traffic visible to them. Owners and admins register, change, test and remove agents, and bind keys to them. To decide which callers may reach which agents across the organisation, use access rules.

The agent gateway is available on plans that include it. Agents with no endpoint are available on every plan. See Billing, plans and credits.