Function calling - H API Documentation

Holo4 (holo4-35b-a3b, holo4-27b) and Holo3.1 (holo3-1-35b-a3b) support this format, so OpenAI-compatible agent frameworks work as is. Pick an output format shows which model takes which. The rest of Core concepts applies here too, including its common pitfalls.

Declare tools

Pass schemas in tools and set tool_choice="required" so the model acts every step. Do not put tool-format examples in the system prompt: the model renders calls in its native format from tools, and a conflicting example degrades quality.

tools = [
    {
        "type": "function",
        "function": {
            "name": "click",
            "description": "Click at (x, y) coordinates",
            "parameters": {
                "type": "object",
                "properties": {
                    "element": {"type": "string", "description": "Detailed description of the target UI element"},
                    "x": {"type": "integer", "description": "X coordinate as integer in [0, 1000]"},
                    "y": {"type": "integer", "description": "Y coordinate as integer in [0, 1000]"},
                },
                "required": ["element", "x", "y"],
            },
        },
    },
    {
        "type": "function",
        "function": {
            "name": "answer",
            "description": "Provide a final answer",
            "parameters": {
                "type": "object",
                "properties": {"content": {"type": "string", "description": "The answer content"}},
                "required": ["content"],
            },
        },
    },
]

const tools = [
  {
    type: "function",
    function: {
      name: "click",
      description: "Click at (x, y) coordinates",
      parameters: {
        type: "object",
        properties: {
          element: { type: "string", description: "Detailed description of the target UI element" },
          x: { type: "integer", description: "X coordinate as integer in [0, 1000]" },
          y: { type: "integer", description: "Y coordinate as integer in [0, 1000]" },
        },
        required: ["element", "x", "y"],
      },
    },
  },
  {
    type: "function",
    function: {
      name: "answer",
      description: "Provide a final answer",
      parameters: {
        type: "object",
        properties: { content: { type: "string", description: "The answer content" } },
        required: ["content"],
      },
    },
  },
] as const;

Loop

Each call in message.tool_calls has a unique id, a function.name, and a JSON string in function.arguments. Plug in your own SYSTEM_PROMPT, observation() (the <observation> message from Core concepts), and execute(name, args).

import json

def run(task: str, max_steps: int = 20) -> str:
    messages = [{"role": "system", "content": SYSTEM_PROMPT}, {"role": "user", "content": task}]
    for _ in range(max_steps):
        messages.append(observation())
        trim_to_last_n_images(messages, n=3)
        resp = client.chat.completions.create(
            model="holo4-27b",
            messages=messages,
            tools=tools,
            tool_choice="required",
            temperature=0.8,
            extra_body={"chat_template_kwargs": {"enable_thinking": True}},
        )
        msg = resp.choices[0].message
        messages.append({"role": "assistant", "content": msg.content, "tool_calls": msg.tool_calls})
        if not msg.tool_calls:
            continue
        for call in msg.tool_calls:
            args = json.loads(call.function.arguments)
            if call.function.name == "answer":
                return args["content"]
            messages.append({"role": "tool", "tool_call_id": call.id, "content": str(execute(call.function.name, args))})
    return "Step budget exhausted"

async function run(task: string, maxSteps = 20): Promise<string> {
  const messages: any[] = [{ role: "system", content: SYSTEM_PROMPT }, { role: "user", content: task }];
  for (let i = 0; i < maxSteps; i++) {
    messages.push(await observation());
    trimToLastNImages(messages, 3);
    const resp = await client.chat.completions.create({
      model: "holo4-27b",
      messages,
      tools,
      tool_choice: "required",
      temperature: 0.8,
      ...({ chat_template_kwargs: { enable_thinking: true } } as any),
    });
    const msg = resp.choices[0].message;
    messages.push({ role: "assistant", content: msg.content, tool_calls: msg.tool_calls });
    if (!msg.tool_calls?.length) continue;
    for (const call of msg.tool_calls) {
      const args = JSON.parse(call.function.arguments);
      if (call.function.name === "answer") return args.content;
      messages.push({ role: "tool", tool_call_id: call.id, content: String(await execute(call.function.name, args)) });
    }
  }
  return "Step budget exhausted";
}

One toolbox can mix screen actions, a code runner, and MCP tools: declare each one in tools.