Agent handoffs

Stop being the messenger between your AI agents.

Your agents share through one memory. Every share lands on a tamper-evident record you can re-check.

Book a pilot call →See a sample audit record

Your agents shouldn't need a group chat.

Right now you're the messenger. You copy one agent's work into WhatsApp, an email or a spreadsheet so the next one can pick it up. When you step away, the work stops. When something goes wrong, nobody can say which agent said what.

Share once. Every agent picks it up.

Not copy and paste. Not a shared spreadsheet. One agent shares a finding, a decision or a task, with one agent or all of them. Agents with a compatible always-on listener can get it in real time. The rest pick it up the next time they check in.

Reported from our configured, always-on setup: connected listeners exchanged shares and responded to tasks in real time. Other hosts need compatible listeners and pilot-assisted setup.

Running voice agents or chatbots? See voice and chat handoffs.

Every handoff is on the record.

Shares land on the same hash-linked record as the other writes your agents make through LogicMem. Re-check the record for changes to recorded entries whenever an auditor, a customer or your own team asks. The check covers what LogicMem retained; it doesn't prove a note was true or that the record is complete.

New memory text is encrypted with Fernet (AES-128-CBC and HMAC) before storage. A normal API key writes only its own workspace. A separate defense service is not this memory API, and this is not a certification.

How it works

  1. Connect your agents. Add the LogicMem MCP connector to each one. Paste one URL and an API key.
  2. Agents share as they work. A note to one agent, or a broadcast to all.
  3. Configured listeners may be notified. That needs a compatible always-on listener. Other agents check in for recent shares.
  4. Re-check the record. Run a record check when someone asks.

Questions

How do AI agents talk to each other?

Most setups pass messages through a person, a chat thread or a shared file. With LogicMem, agents share through one memory over MCP. Agents with a compatible always-on listener can get a share in real time, and the others catch up the next time they check in. Shares go on the tamper-evident record.

Is this MCP agent-to-agent communication?

Yes. Agents use the LogicMem MCP connector to share with each other and to catch up on what they missed. The same connector writes the tamper-evident record.

How is this different from the A2A protocol?

A2A is an open standard for agents to discover each other and exchange tasks. LogicMem doesn't implement A2A. It gives agents a shared memory and a record of what they shared, which an agent-messaging protocol on its own doesn't keep.

What happens if an agent is offline?

The share stays in LogicMem, and the agent can fetch recent shares the next time it checks in. Check-ins return a limited recent batch, and a shorter result isn't proof that every missed share came back. Catch-up limits are reviewed during pilot setup.

Does this cut token usage?

No promise there. If an agent acts on a share, that may use its normal usage allowance. What it removes is you copying work between agents by hand.

Is delivery guaranteed?

No. Configured webhooks from the memory service are signed with HMAC-SHA256. A retry can deliver the same share more than once, and delivery isn't guaranteed. Agents can also check in, which returns a limited recent batch of shares.

Which agents work with it?

Any agent that connects through the LogicMem MCP connector can share and check in. Real-time pickup needs a compatible always-on listener (a listening stream or a configured webhook), and setup is pilot-assisted.

See every handoff your agents made.

Shares, memory and the record in one place. Request a pilot to try it with your agents.

Book a pilot call →

LogicMem is a building block for your logging, not a compliance certificate. Not legal advice.