Model Provider Request★

A model provider request is the structured network payload (usually a JSON POST request) that a developer's harness sends to an LLM API endpoint (like Anthropic or OpenAI) to generate a completion.

You configure this request whenever you:

• Write code to call Claude or GPT from your server backend.
• Define system prompts and select a temperature parameter.
• Pass a list of functions (tools) that your agent is allowed to invoke.

Anatomy of a Request Payload

A standard request is not just a raw string of text. It is a structured JSON object containing several configuration blocks:

1. Model Identifier: Tells the provider which specific parameters to load (e.g. claude-3-5-sonnet-20241022).
2. System Prompt: The root instructions that set the agent's behavior and boundaries.
3. Messages Array: The dialogue history, structured as objects with role (user, assistant, or system) and content fields.
4. Parameters: Tuning options like temperature, max_tokens, and response_format.
5. Tools (Schemas): Descriptions of functions the model can call, formatted as JSON Schemas.

+-----------------------------------------------------------+
|               MODEL PROVIDER REQUEST PAYLOAD              |
|                                                           |
|  [Model ID] ────► "claude-3-5-sonnet"                     |
|  [System]   ────► "You are an SQL generator..."           |
|  [Messages] ────► [{"role": "user", "content": "Query"}]  |
|  [Tools]    ────► [{"name": "run_sql", "parameters":{}}]   |
+-----------------------------------------------------------+

Field Applications

1. Fullstack & AI Engineers (Standard API Call)

Engineers construct this request object using the provider's SDK library:

• Code Example:

  const response = await client.messages.create({
    model: "claude-3-5-sonnet",
    max_tokens: 1000,
    temperature: 0.2,
    system: "You are a compiler assistant.",
    messages: [
      { role: "user", content: "Explain this type error: ..." }
    ],
    tools: [{
      name: "get_type_definition",
      description: "Fetch type information for a TS interface.",
      input_schema: { type: "object", properties: { name: { type: "string" } } }
    }]
  });