How Do AI Agents on Different Computers Talk to Each Other? I Built a Tiny Open-Source Fix

Description and use

ACP‑lite enables AI agents on separate home computers behind NAT to securely exchange messages without a central server, using Syncthing file sync and SSH wake‑up.

Problem

Running multiple AI agents on separate home computers behind NAT without a public IP or server makes inter-agent communication difficult; existing enterprise solutions are too heavyweight.

Outcome

ACP‑lite enables secure, token‑free, end‑to‑end signed communication between AI agents on two home computers behind NAT, with minimal setup and no central server.

Workflow

• 1. Each machine runs a local "post office" process that watches an outbox folder.
• 2. An agent writes a JSON letter to its local outbox.
• 3. Syncthing syncs the letter file to the other machine’s inbox.
• 4. The receiving machine’s watch process detects the new file and triggers an SSH "doorbell" to wake the target agent.
• 5. The agent reads the letter, processes the task, writes artifacts and a signed reply to its outbox.
• 6. The audit log chain verifies end‑to‑end integrity.
• 7. The process repeats for subsequent messages.

Evidence

I run 6 AI agents across 3 machines: Codex and DSH on my desktop, Hermes and OpenClaw on a netbook. Human-to-agent is easy — WeChat, DingTalk, done. But how do agents talk to each other?

Source

each machine runs a "post office"; agents only ever read and write local files.

Source

Syncthing syncs the file to the other machine — battle-tested decentralized file sync, no need to reinvent it

Source

SSH is the doorbell, ACP is the mailbox; the doorbell carries only a pointer, the letter body always travels via ACP

Source

The completion criteria are threefold, all required: artifact on disk, longer audit chain, signed receipt.

Source