Agent Configuration
Agents are the AI backends that process your messages. ClawCenter supports nine types.
Agent Types
Claude Code (CLI)
Spawns a local claude CLI process. Requires the Claude Code CLI installed.
| Field | Description |
|---|---|
| Working Directory | The project folder where Claude Code runs |
| Model | Model to use (optional) |
The CLI runs with --dangerously-skip-permissions for non-interactive use.
Claude Agent SDK
Calls the Anthropic API directly via @anthropic-ai/claude-agent-sdk. No CLI installation needed, but requires an ANTHROPIC_AUTH_TOKEN environment variable.
| Field | Description |
|---|---|
| Working Directory | The project folder context |
| Model | Model name (default: sonnet) |
OpenCode (CLI)
Spawns a local opencode CLI process. Requires OpenCode installed.
| Field | Description |
|---|---|
| Working Directory | The project folder where OpenCode runs |
| Model | Model to use (set via OPENCODE_MODEL env var) |
OpenClaw (CLI)
Spawns a local openclaw CLI process. Requires OpenClaw installed.
| Field | Description |
|---|---|
| Working Directory | The project folder where OpenClaw runs |
| Model | Model to use (optional) |
Codex (CLI)
Spawns a local codex CLI process. Requires OpenAI Codex CLI installed.
| Field | Description |
|---|---|
| Working Directory | The project folder where Codex runs |
| Model | Model to use (optional) |
| Permission Mode | Permission level (optional) |
CodeBuddy (CLI)
Spawns a local codebuddy CLI process. Requires Tencent CodeBuddy CLI installed.
| Field | Description |
|---|---|
| Working Directory | The project folder where CodeBuddy runs |
| Model | Model to use (optional) |
| Permission Mode | Permission level (optional) |
Cursor Agent (CLI)
Spawns a local cursor CLI agent process. Requires the Cursor CLI installed.
| Field | Description |
|---|---|
| Working Directory | The project folder where Cursor Agent runs |
| Model | Model to use (optional) |
| Permission Mode | Permission level (optional) |
Custom HTTP
Sends messages to any HTTP endpoint. Use this to integrate your own AI services.
| Field | Description |
|---|---|
| URL | The HTTP endpoint (e.g. http://localhost:3000/chat) |
| Headers | Custom headers (optional) |
The endpoint should accept POST with { message, session_id } and return { text }.
Remote Worker
Agents running on remote machines connected via WebSocket. See Multi-Machine Deployment. These agents are registered automatically when a Worker connects — you don't create them manually.
Multiple Agents of the Same Type
You can create multiple agents of the same type with different working directories. This is the recommended way to work with multiple projects:
| ID | Type | Working Directory |
|---|---|---|
frontend | Claude Code | /home/user/frontend |
backend | Claude Code | /home/user/backend |
infra | Claude Code | /home/user/infra |
Each agent has its own process, sessions, and conversation history.
Starting and Stopping
- Agents can be started/stopped individually from the Web UI
- When ClawCenter starts, it automatically starts all previously running agents
- Stopped agents are not available for routing — messages to a stopped agent return an error
Agent Display Name
The display name appears in reply prefixes (e.g. [🤖 Claude]). Use emoji and short names to make it easy to identify agents in WeChat.