# AI Coding Tools

Termbridge works great for monitoring AI coding assistants. Start an agent in your terminal, then watch it work from your phone while you're away from your desk.

## Why use Termbridge with AI tools?

- **Monitor remotely**: Watch Claude Code or Codex work from your couch
- **Stay informed**: See when agents need input or hit errors
- **Quick intervention**: Send commands or cancel processes from your phone
- **No VPN needed**: Secure access from anywhere

## Supported tools

Termbridge streams whatever runs in the tmux session, so it doesn't need to know about the tool. The examples on this page use:

- **Claude Code** (`claude`)
- **Codex** (`codex`)
- **OpenCode** (`opencode`)

These three are also the agents that sandbox mode can install and sign in automatically. Other terminal-based agents run the same way but are not tested.

## Local workflow

### Watch Claude Code from your phone

```bash
# Start Termbridge with a named session
termbridge --session claude

# In the terminal (via UI or locally), start Claude
claude
```

Scan the QR code and watch Claude work from anywhere.

### Start the agent first

If you prefer to start the agent before Termbridge:

```bash
# Create a tmux session and start the agent
tmux new-session -d -s claude "claude"

# Then attach Termbridge to it
termbridge --session claude
```

### Codex workflow

```bash
tmux new-session -d -s codex
tmux send-keys -t codex "codex" Enter
termbridge --session codex
```

## Sandbox workflow

For isolated environments, run agents in a Daytona sandbox. Termbridge can install agent CLIs and sync your auth automatically.

### Quick setup

```bash
export TERMBRIDGE_BACKEND=sandbox-daytona
export TERMBRIDGE_SANDBOX_AGENTS=claude-code,codex

termbridge
```

This will:

1. Create a Daytona sandbox
2. Install the agent CLIs
3. Sync your local auth files so agents are ready to use

See the [Coding Agents guide](/docs/sandboxes/coding-agents) for full details.

## Tips

### Named sessions

Use meaningful session names so you know what's running:

```bash
termbridge --session claude-refactor
termbridge --session codex-tests
```

### Multiple agents

Run different agents in different terminals:

```bash
TERMBRIDGE_SESSIONS=2 termbridge
```

Start Claude in one, Codex in another, and switch between them.

### Long-running tasks

AI agents can run for a long time. Termbridge handles this well:

- Your session persists even if you close the browser
- Reconnect anytime by refreshing the page
- tmux keeps the agent running even if Termbridge restarts

### Intervention

Need to stop or redirect an agent?

1. Use the `Ctrl+C` button to interrupt
2. Type new instructions in the input field
3. The agent receives your input just like a local terminal

## Sandbox vs local

| Feature | Local (tmux) | Sandbox (Daytona) |
|---------|--------------|-------------------|
| Setup | Just tmux | Daytona account + API key |
| Environment | Your machine | Isolated container |
| Persistence | Survives restarts | Survives laptop sleep in direct mode |
| Auth | Already configured | Auto-synced by Termbridge |
| Best for | Quick monitoring | Isolated experiments |

For most users, local mode is simpler. Use sandbox mode when you want:

- A clean, disposable environment
- To keep your laptop free while agents work
- To run experiments without affecting your local setup