Agents

Agents are AI tools which do a specific task in your application. The Django AI Core Agent Framework allows you to register agents which can then either be called directly in your application code, or via a URL on your frontend.

Future plans

Right now agents aren't much different from views, but the hope is that eventually by registering an agent you will get benefits like:

• Logging
• Rate limiting
• Auth
• Streaming responses

Registering an Agent

Register your agents in your application using the Agent registry:

# agents.py
from django_ai_core.contrib.agents import Agent, AgentParameter, registry
from django_ai_core.llm import LLMService


@registry.register()
class SimplePromptAgent(Agent):
    slug = "prompt"
    name = "Basic agent"
    description = "Basic agent that just takes a prompt and returns a response from the LLM."
    parameters = [
        AgentParameter(
            name="prompt",
            type=str,
            description="The prompt to use for the agent",
        ),
    ]

    def execute(self, *, prompt: str):
        service = LLMService.create(provider="openai", model="gpt-4o")
        response = service.completion(prompt)

These can be in any file you want, but if you use a file like agents.py, you'll need to make sure Django loads it in your apps.py:

from django.apps import AppConfig


class MyAppAppConfig(AppConfig):
    label = "myapp"
    name = "mayapp"
    verbose_name = "My example app"

    def ready(self):
        from . import agents

Configuring Agent URLs

Agents can be made accessible at automatically generated URLs by adding agent_urls() to your urlpatterns:

from django.urls import path, include
from django_ai_core.contrib.agents.urls import agent_urls as ai_core_agent_urls

urlpatterns = [
    ...
    path("ai/", include(ai_core_agent_urls())),
]

This will generate URLs under ai/ for all your registered agents based on their slug values. The SimplePromptAgent from the above example will be accessible at ai/prompt/.

Alternatively, you can register views for individual agents using the as_view classmethod on Agent classes:

from django.urls import path, include
from .agents import SimplePromptAgent

urlpatterns = [
    ...
    path("ai/simple/", SimplePromptAgent.as_view()),
]

## Using Agents

Agents can either be invoked directly:

```python
from .agents import SimplePromptAgent

SimplePromptAgent().execute(prompt="Foo")

or via their URL:

POST https://www.example.com/ai/prompt/

{
    "arguments": {
        "prompt": "Are you suggesting coconuts migrate?"
    }
}

Permissions

You can control who can execute your agents using permission. See the Permissions documentation for detailed information on how to secure your agents.

Quick example:

from django_ai_core.contrib.agents import Agent
from django_ai_core.contrib.agents.permissions import IsAuthenticated

@registry.register()
class SecureAgent(Agent):
    slug = "secure"
    name = "Secure Agent"
    description = "Agent requiring authentication"
    permission = IsAuthenticated()
    parameters = []

    def execute(self):
        return "Authenticated access only!"