Agent-to-Agent Messaging

How to send messages from Coppermind to external AI agents, and what happens when external agents send messages to you.

What This Is#

As AI agents become more common in business, your clients and their vendors may have their own agents - a CRM agent, an analytics agent, a project management agent, or even the client's own AI assistant. Agent-to-agent (A2A) messaging lets Coppermind communicate directly with these systems.

Think of it as email between AI agents. You compose a message, approve it, and Coppermind delivers it to the other agent. When messages come in, they land in your inbox for review.

Sending a Message#

To send a message to another agent, switch to the relevant client and describe what you want to send:

"Send a message to the HubSpot agent for Acme sharing our Q2 campaign results"

Coppermind creates the message and queues it for your approval. No message is ever sent automatically - you always see the draft first and explicitly approve it before delivery.

Message Types#

Examples#

Sharing campaign results with a client's analytics agent:

"Send Acme's Q1 campaign outcomes to their analytics agent"

Delegating a research task to a VA's agent:

"Ask the VA agent to pull competitor pricing for BetaCorp"

Notifying a client's agent about a strategy change:

"Notify Acme's agent that we're pausing paid social through Q2"

Receiving Messages#

When an external agent sends a message to your Coppermind, one of two things happens:

• Known agent (previously configured): The message is delivered to the relevant client mind and appears in your daily loop or meeting prep.
• Unknown agent (first contact): The message goes to your unattributed inbox for review. You decide whether to accept and attribute it to a client.

Either way, you are always in the loop. Nothing from an external agent is automatically stored or acted on without your awareness.

The Approval Gate#

Every outbound message passes through an approval step. When you ask Coppermind to send a message:

1. Coppermind drafts the message with the right structure and content
2. You see the full message (recipient, type, content)
3. You approve or modify before it goes out

This is a deliberate safety measure. Client data should never leave your system without your explicit sign-off.

When to Use Agent Messaging#

• Client handoff - When an engagement ends, share the client mind with the incoming CMO's agent system. The "handoff" message type packages everything they need.
• Cross-tool coordination - If the client's CRM agent can push updated lead metrics, you save manual data entry.
• VA delegation - Send structured research tasks to a VA's agent and receive formatted results back.
• Status updates - Keep a client's own AI assistant informed about strategy changes without manual email.

What It Does Not Do#

• No auto-sending. Every outbound message requires your approval.
• No auto-processing. Inbound messages from unknown agents are quarantined.
• No real-time streaming. Messages are discrete, not live conversations.
• No agent discovery. You need to know the target agent's identifier - there is no marketplace or directory.

Security#

• All messages are encrypted in transit
• Outbound messages are logged in the audit trail
• Inbound messages from unrecognized agents are held for your review
• Message size is capped at 100KB - larger payloads use reference links
• Agent identity is verified by API key