Body Builder

Not yet available.

Body Builder is a request preprocessing layer that enhances your prompts before they reach the model. Based on the detected task — coding, writing, structured extraction, multilingual chat — it injects a system prompt, formatting constraints, and output instructions tuned for that task. You send a minimal request; Body Builder fills in the scaffolding.

What it does#

When Body Builder is enabled for a request, OpenDunes:

1. Classifies the task from the user message
2. Selects the appropriate system prompt template for that task type
3. Injects the template as the system message (or prepends to an existing one)
4. Optionally adds output formatting constraints (json_schema, response length guidance, etc.)

The model receives a fully constructed prompt. Your code only sends the raw user message.

Enabling Body Builder#

Pass "body_builder": true (or a configuration object) in the request body:

{
  "model": "anthropic/claude-sonnet-5",
  "body_builder": true,
  "messages": [{ "role": "user", "content": "Debug this Python function: def add(a, b): return a - b" }]
}

With true, Body Builder detects the task automatically. OpenDunes detects this as a coding/debugging task and injects an appropriate system prompt focused on code review and corrections.

Configuration object#

For finer control, pass a configuration object instead of true:

{
  "model": "anthropic/claude-sonnet-5",
  "body_builder": {
    "task": "coding",
    "language": "python",
    "output_format": "diff",
    "inject_system": true,
    "append_to_existing_system": false
  },
  "messages": [{ "role": "user", "content": "Refactor this class to use dataclasses." }]
}

Configuration fields#

Transparency#

The constructed system prompt is visible in the response under body_builder.injected_system when debug mode is on:

{
  "body_builder": {
    "task_detected": "coding",
    "injected_system": "You are an expert Python developer..."
  }
}

This field is only present when "debug": true is included in the request.

Supported task types and templates#

Pricing (DA)#

Body Builder's injected tokens (system prompt) count as prompt tokens and are billed at the model's input DA rate. Injected system prompts are typically 50–200 tokens. This adds a small incremental cost per request.

Note. Body Builder is a convenience layer, not a jailbreak or safety mechanism. It injects helpful scaffolding — it does not enforce content policies beyond what the model already applies.