Skip to content

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.

FieldDescription
Working DirectoryThe project folder where Claude Code runs
ModelModel 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.

FieldDescription
Working DirectoryThe project folder context
ModelModel name (default: sonnet)

OpenCode (CLI)

Spawns a local opencode CLI process. Requires OpenCode installed.

FieldDescription
Working DirectoryThe project folder where OpenCode runs
ModelModel to use (set via OPENCODE_MODEL env var)

OpenClaw (CLI)

Spawns a local openclaw CLI process. Requires OpenClaw installed.

FieldDescription
Working DirectoryThe project folder where OpenClaw runs
ModelModel to use (optional)

Codex (CLI)

Spawns a local codex CLI process. Requires OpenAI Codex CLI installed.

FieldDescription
Working DirectoryThe project folder where Codex runs
ModelModel to use (optional)
Permission ModePermission level (optional)

CodeBuddy (CLI)

Spawns a local codebuddy CLI process. Requires Tencent CodeBuddy CLI installed.

FieldDescription
Working DirectoryThe project folder where CodeBuddy runs
ModelModel to use (optional)
Permission ModePermission level (optional)

Cursor Agent (CLI)

Spawns a local cursor CLI agent process. Requires the Cursor CLI installed.

FieldDescription
Working DirectoryThe project folder where Cursor Agent runs
ModelModel to use (optional)
Permission ModePermission level (optional)

Custom HTTP

Sends messages to any HTTP endpoint. Use this to integrate your own AI services.

FieldDescription
URLThe HTTP endpoint (e.g. http://localhost:3000/chat)
HeadersCustom 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:

IDTypeWorking Directory
frontendClaude Code/home/user/frontend
backendClaude Code/home/user/backend
infraClaude 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.

Released under the Apache 2.0 License.