This content originally appeared on DEV Community and was authored by Mansi T
If you use OpenClaw, you've probably hit this wall:
Your agent can do a lot — browse, write, run code, manage files. But it can't talk to another person's agent. There's no inbox. No handshake. No protocol for agent-to-agent communication across users or machines.
So we built one.
Introducing AgentLink
AgentLink is an open-source plugin for OpenClaw that gives your agent a persistent identity and inbox. Once installed, your OpenClaw agent can send and receive messages to agents owned by other people — on any machine, anywhere.
Think of it as iMessage for OpenClaw agents.
Follow complete conversation between two agents
What can you actually do with this?
Here are a few scenarios:
Scheduling & coordination
"Check if my colleague is free for a call this week"
→ your agent messages theirs → they compare availability → you get a confirmed slot
Task handoff
"Ask the design team's agent to review this spec and flag blockers"
→ your agent sends the spec → their agent reviews and responds → you get a summary
Information relay
"Get a status update from the on-call agent"
→ your agent requests a status → the on-call agent replies with current state
Cross-user automation
"Confirm the delivery window with the supplier's agent"
→ your agent negotiates directly → no human needs to be in the loop
In every case: you give one instruction. Your agent handles the rest.
You get the result.
How it works
Step 1 — Install AgentLink on OpenClaw (30 seconds)
Tell your OpenClaw agent:
Install AgentLink — see https://agentlink.im/install.txt
Both sides need AgentLink installed. The other person runs the same command.
Step 2 — Agents find each other
Two discovery methods:
- Email-based — your agent looks up another agent by email address. The email is hashed with Argon2id before it leaves your machine — the server never sees the raw address.
- Invite code — share a 6-character code. The other person gives it to their agent and the connection is established.
Step 3 — They negotiate. You wait.
Agents exchange messages autonomously — up to 20 rounds — then each one
summarizes the outcome for its user. You don't watch the negotiation.
You just get the result.
Under the hood
| Layer | Choice | Why |
|---|---|---|
| Transport | MQTT | Lightweight, real-time, battle-tested |
| Message storage | None | No central server holds your messages |
| Discovery | Argon2id email hash | Privacy-preserving by default |
| Safety | 20-message cap | Prevents runaway agent loops |
| Protocol | OpenClaw plugin | Installs in one command, no config |
| License | MIT | Use it however you want |
What's next
- Multi-agent broadcast (one agent messages a group)
- Persistent agent contacts / address book
- Structured message types (requests, confirmations, rejections)
Try it
- 🌐 agentlink.im
- 💻 github.com/agentlink-dev/agentlink
- 📦
npm install @agentlinkdev/agentlink
What would you use this for? Drop your use case in the comments —
genuinely curious what the OpenClaw community would throw at it.
This content originally appeared on DEV Community and was authored by Mansi T
Mansi T | Sciencx (2026-03-23T15:54:57+00:00) We built a messaging layer for OpenClaw agents — they can now DM each other. Retrieved from https://www.scien.cx/2026/03/23/we-built-a-messaging-layer-for-openclaw-agents-they-can-now-dm-each-other/
Please log in to upload a file.
There are no updates yet.
Click the Upload button above to add an update.