Use application tools

An application orchestrator connecting a language model to calculator, database, clock, and action tools

Use tools for calculations, current information, internal records, and actions.

Separate SERV Tools from application tools

Application tools describe functions that your code executes. SERV Tools are control markers that SERV consumes before the model runs. SERV recognizes a fixed set of marker names; use the documented names for SERV features. You can put both kinds in one tools array. SERV removes its marker tools and forwards your application tools unchanged. The model can call get_order, but it never calls serv_prompt_guard or serv_shadow_agent.

const response = await client.chat.completions.create({
  model: "gpt-5.4-mini",
  messages: [
    { role: "system", content: "Use get_order for order status. Never guess." },
    { role: "user", content: "What is the status of order 1842?" },
  ],
  tools: [{
    type: "function",
    function: {
      name: "get_order",
      description: "Look up an order by its ID.",
      parameters: {
        type: "object",
        properties: { order_id: { type: "string" } },
        required: ["order_id"],
        additionalProperties: false,
      },
    },
  }],
});

Your application executes the function call and sends its result in a follow-up request. SERV removes its recognized marker tools; your application tools are forwarded normally. An unrecognized tool name is treated as an application tool, even if it begins with serv_. Describe when a tool should be used, define a strict input schema, and validate the arguments before performing side effects. The system prompt should state the tool-use policy, for example: “Use get_order for every order-status question. Never guess an order status.”

Combine application tools with SERV Tools

The same request can enable SERV protection, quality validation, and an application function:

tools: [
  { type: "function", function: { name: "serv_prompt_guard" } },
  {
    type: "function",
    function: {
      name: "serv_shadow_agent",
      parameters: {
        type: "object",
        properties: { hint: { type: "string", default: "Cite the order status returned by the tool." } },
      },
    },
  },
  {
    type: "function",
    function: {
      name: "get_order",
      description: "Look up an order by its ID.",
      parameters: {
        type: "object",
        properties: { order_id: { type: "string" } },
        required: ["order_id"],
        additionalProperties: false,
      },
    },
  },
]

SERV consumes the first two markers. Only get_order reaches the model. Your server remains responsible for executing get_order, checking the arguments, and handling the follow-up conversation.

Info

API reference

Guides

Research

Changelog

Separate SERV Tools from application tools

Combine application tools with SERV Tools

​Separate SERV Tools from application tools

​Combine application tools with SERV Tools

Separate SERV Tools from application tools

Combine application tools with SERV Tools