Getting Started

Prerequisites

• Python 3.10+, Django 4.2+, Django REST Framework 3.14+.

1. Install

pip install drf-llm-gateway

Add "drf_llm_gateway" to INSTALLED_APPS only if you want the generate_llm_tools management command (Django discovers management commands per app). Every other feature works without it.

INSTALLED_APPS = [
    ...,
    "drf_llm_gateway",  # optional - only needed for the CLI
]

2. Register a viewset

from rest_framework import viewsets
from drf_llm_gateway import expose_as_tool


@expose_as_tool(actions=["list", "retrieve", "create"])
class ArticleViewSet(viewsets.ModelViewSet):
    queryset = Article.objects.all()
    serializer_class = ArticleSerializer

This registers three tools — article_list, article_retrieve, article_create — into the module-level default registry, the moment this module is imported (which happens naturally once a router imports it in your urls.py).

3. Export the tools

from drf_llm_gateway import to_openai_tools

tools = to_openai_tools()

[
  {
    "type": "function",
    "function": {
      "name": "article_list",
      "description": "List Article objects.",
      "parameters": {"type": "object", "properties": {}}
    }
  },
  {
    "type": "function",
    "function": {
      "name": "article_retrieve",
      "description": "Retrieve a single Article by its identifier.",
      "parameters": {
        "type": "object",
        "properties": {"pk": {"type": "integer"}},
        "required": ["pk"]
      }
    }
  }
]

Pass tools directly as the tools= argument of an OpenAI Chat Completions or Responses API call.

4. Execute a tool call

When the model responds with a tool call, dispatch it with execute_tool:

from drf_llm_gateway import execute_tool

result = execute_tool(
    "article_create",
    {"title": "Hello World", "author": 1},
    user=request.user,
)
print(result.status_code, result.data)

This runs through the real ArticleViewSet — the same authentication, permission checks, and serializer validation a genuine HTTP POST would go through — so there is no separate, unvalidated code path for LLM-initiated calls.

Next steps

• Quick Start — the full range of field types supported, custom actions, examples in the schema.
• Advanced Usage — MCP export, OpenAPI-adjacent concerns, permission modeling for agent identities.
• Security — read this before exposing tools to an agent that isn't fully trusted.