imux · 龍imux
AI workbench for macOS

Function calling & tools

Like docs.x.ai tools guides, Clavue supports tool use on three layers: (1) your app’s OpenAI tools schema on the API, (2) Official MCP hosted tools, (3) imux Agent Chat native tools (filesystem, browser, shell under workspace policy).

OpenAI tools on the API

Send tools / tool_choice on chat completions when the product model supports tool use. Your client executes tool calls and returns role: tool messages.

{
  "model": "clavue-2.1",
  "messages": [{"role": "user", "content": "What's the weather in Tokyo?"}],
  "tools": [{
    "type": "function",
    "function": {
      "name": "get_weather",
      "description": "Get current weather by city",
      "parameters": {
        "type": "object",
        "properties": {"city": {"type": "string"}},
        "required": ["city"]
      }
    }
  }]
}

Official MCP · web_search

Hosted search is not free-form chat. Use model id clavue-2.1-search or the MCP tool mcp__clavue__web_search. Output is agent-normalized JSON with sources.

imux native tools

In Agent mode, imux exposes read/write/search/browser/shell tools bound to the workspace root. Diffs are reviewable before apply. Non-focus socket automation must not steal UI focus.

  • Prefer @file / @symbol context over dumping whole monorepos
  • Use Plan mode for read-only exploration
  • Mission Control for multi-agent fleet + verify predicates

Next