Devices & other agents
Pair your phone and other computers, let one Suzent drive another, delegate to A2A agents, and use Suzent from your editor.
Sovereign vessel. Stretching an agent across devices is where most systems quietly hand control to a cloud relay. Suzent connects only the devices you pair, over links you approve, with each device declaring what it can do. See what makes an agent sovereign.
Suzent can reach beyond the computer it runs on:
- Your phone can chat with your agent through the Suzent app.
- Other devices (laptops, servers, phones, even a TV browser) can lend the agent their hardware, such as a camera or speaker.
- Another Suzent on a different computer can be asked to do work with its own files and memory.
- Other agents that speak the open A2A standard can take on delegated work.
- Your code editor can use Suzent as its agent through ACP.
Everything here is managed in Settings → Devices and Settings → Mesh.
Step 1: Make this computer reachable
Out of the box, Suzent only listens to your own computer, so other devices can't reach it. To change that, open Settings → Mesh → Network access, turn it on, and restart Suzent. Your other devices can then find it on your home network or your Tailscale network.
Being reachable doesn't open anything up by itself: every device still has to be approved by you, and an approved device can only do what you granted.
warning
Only do this on a network you trust, such as your home Wi-Fi or Tailscale. Letting another device drive your agent is close to letting it run code on your computer.
Pair your phone
The Suzent app for Android and iOS pairs from Settings → Devices → Mobile access, with its own permissions for each phone. See Mobile app.
Connect another device
Open Settings → Devices → Discover. Suzent lists devices on your local network and your Tailscale network.
-
To use a device's hardware, click Connect on it. A pairing code appears, and whoever owns that device approves it in their Settings → Devices.
-
To add a phone, tablet, or TV without installing anything, use Add a screen (phone, tablet, TV) and scan the QR code with that device's camera.
-
To connect a headless server, run this on the server, then approve the pairing code on your main computer:
suzent nodes host --name "My Server"
Approval is a one-time step. The device remembers its access and reconnects on its own until you revoke it.
Once connected, just ask: "take a photo with my phone", or "play this on the living-room speaker". You can see each device's abilities in the device list.
A computer's webcam is offered only once the Camera plugin is installed on it.
Drive another Suzent
If both computers run Suzent, one can ask the other's agent to do work using that computer's own files, memory, and tools.
- On the computer that will give the orders, open Settings → Devices → Discover and click Control on the other one.
- On the other computer, approve the request in Settings → Devices.
- Now ask your agent, for example, "ask my desktop to summarize ~/notes". The other computer's reply streams back, and the conversation appears in its chat list too.
From the command line, with files attached:
suzent nodes trigger "Desktop" "summarize these" --file ./notes.txtA control grant lets the other computer run its agent for you, and nothing else: it can't read your settings or files directly. Each link has two independent directions, so "I can control them" and "they can control me" are granted and revoked separately. Pause or remove a link at any time in the device list.
To operate a computer fully from elsewhere, create a host token in Settings → Security. It grants complete access, is shown only once, and can be revoked like any device. Treat it like a password.
Work with other agents (A2A)
A2A is an open standard that lets agents built by different people delegate work to each other.
- To delegate to an external agent, add its address under Settings → Mesh → Agents in your mesh. Your agent can then hand it tasks, and you can follow them under Delegated work.
- To let other agents find this device, turn on This device in Settings → Mesh. That publishes a card describing your agent. Publishing is off by default, and it grants nothing: another agent still needs a grant you approve before it can send work. Work sent to you appears under Work given to you.
Use Suzent from your editor (ACP)
Editors and tools that support the Agent Client Protocol, such as Zed, can use Suzent as their AI agent. Each session is a real Suzent conversation, with your memory, skills, models, and permission rules, and it shows up in the desktop app too.
- Keep Suzent running (
suzent start, or the background service). - In your editor's settings for external agents, add a custom agent with the
command
suzentand the argumentacp. - Choose Pair with Suzent when your editor asks you to sign in. In Settings → ACP Agents → Editor connections, compare the code with the login terminal, check the confirmation box, and approve the request. Return to the editor and reconnect after login succeeds.
Editors without terminal login can choose Approve in Suzent settings;
the code appears in their agent log. Alternatively run
suzent acp --login --client-name "<editor name>" yourself, using the name
reported by the editor, then reconnect. Requests expire after five minutes.
Each authorization can resume only the conversations it created. It uses your Suzent memory, skills and workspace tools, and the editor handles tool approval prompts. Only approve editors you trust. Revoke a connection from the same settings section to block further requests and stop active work. A new login creates a new authorization; it does not recover the old authorization's chats. Credentials are saved per backend address and editor name, so instances using the same name and address share an authorization. Changing either requires login again. Existing unpaired ACP conversations remain available in Suzent itself.
For existing trusted local integrations, suzent acp --no-require-auth explicitly
opts out of pairing. This also opts out of authorization-based chat isolation.
The editor's open folder becomes the agent's working folder. When the agent
needs permission, the editor asks you. To skip those prompts for unattended
use, add --permission-mode auto or --permission-mode full_access to the
arguments.
It also works the other way around: under Settings → ACP Agents, Suzent can use coding agents such as Claude Code or Codex as helpers.
Opening the page or clicking Refresh checks installed agents automatically. An agent that passes the check shows Signed in, with Sign in again available if you need to repeat authorization. Installation alone is not a login check.
To sign in, click Sign in beside an installed agent. Finish the agent's login flow in the browser or terminal that opens on the computer running Suzent. The page shows progress, lets you cancel, and checks the connection again when login finishes. If the agent offers several methods, use Probe to discover them and choose a method before signing in. Once it reports Reachable, retry your task. Suzent keeps the saved conversation ID when authentication expires; it does not start a replacement conversation just because login is required.
Signing in from this page requires access from the same computer. On a host without a graphical terminal, expand Details to copy the login command, run it on that host, then probe again. Installation and login help stays collapsed until needed. Installation status alone does not confirm authentication, and a successful probe confirms session creation rather than availability of every model or tool.
Command line reference
suzent nodes discover # find devices on your network and Tailscale
suzent nodes list # connected devices
suzent nodes describe <name> # what a device can do
suzent nodes pending # pairing codes waiting for approval
suzent nodes approve <code>
suzent nodes devices # approved devices
suzent nodes revoke <device_id> # remove a device; it must pair again