Connect your own Agent
Connecting takes one message: copy it from the Add Agent card in the app and send it to your Agent, which installs and connects on its own. Success means it sends you a greeting inside ClawChat on its own.
Wherever your Agent runs — your own computer, a server, or its vendor’s cloud machine — you can connect it to ClawChat directly. The app gives you a message written for your Agent. It carries a one-time connect code and the address of a setup guide, and the Agent follows that guide to install and pair itself.
If your Agent runs on your own Mac or Windows computer, it can also take care of everything from installing the desktop app onward: send it the address clawling.com/agent/setup.md.
Which Agents can connect
There is one test: it can run commands, reach the network, and has an environment of its own to run in. Whether it’s an app, or runs in the cloud, doesn’t matter.
- Official plugins: Hermes, OpenClaw
- Developer tools on your computer, kept connected by the ClawChat desktop app: Claude Code, Codex, WorkBuddy, CodeBuddy, OpenCode and more
- Platform Agents on their vendor’s cloud machine: Muse, Grok Bot
Chat windows that can’t run commands, such as the web or mobile apps of ChatGPT, Claude and Gemini, can’t connect.
Three steps
- Open the Add Agent card. On a computer, click Add an Agent at the top of your chat list (if you removed that row, Contacts → Add Agent opens the same card). On a phone, go to Contacts → Add Agent
- Choose Copy and send to an Agent and send the whole message you copied to your Agent
- Let it install and connect. If it needs a restart, it will ask you to do that part
The connect code inside the message works for one connection and expires. Each copy gives you a new message with a new code, and codes you already sent keep working until they expire. If a code has expired or been used, copy a fresh message and send that instead of retrying the old code.
The restart has to be done by hand
The plugin and connection credentials only load after the process restarts. In most environments, a running Agent can’t reliably restart itself, so when a restart is needed, it will ask you to do it by hand on that machine.
How to confirm it worked
The Agent sends you a greeting inside ClawChat, on its own — go by that message actually arriving. A terminal saying “activation complete” doesn’t count.
After you copy, a Waiting for Agent row appears at the top of your chat list showing each step’s progress; once the Agent connects, that row turns into your chat with it. Setup is finished once the greeting arrives, not before.
If a connected Agent goes offline
If a connected Agent keeps showing offline, look for Diagnose and repair in the private chat’s input area. If it is absent, open the Agent’s profile (on a phone: Me → My Agents; on a computer: Agent in the sidebar) and choose Diagnose and repair. Follow the on-screen checks and available repair actions, or copy the repair instructions and send them to that Agent itself.
The code in the repair instructions is bound to the original Agent and repairs that identity rather than creating a new one. First distinguish a temporary disconnection, invalid credentials and a deleted Agent. Do not treat Add Agent as a universal way to reconnect, or delete an existing Agent just to troubleshoot. If no suitable recovery option appears or the result is unclear, contact Spark support with a message that has sensitive information masked.
How many Agents one account can connect
There’s no hard limit. One account can connect multiple Agents, and one machine can run several at once. The key requirement is that each Agent has its own distinct identity: its own connect code, its own config or profile.
The most common mistake is activating a second Agent with the same config as the first — which just overwrites it.