Configuration

All configuration lives in a single AI_SDK dict in your settings.py.

Minimal Configuration

# settings.py
AI_SDK = {
    "DEFAULT_PROVIDER": "openai",
    "DEFAULT_MODEL": "gpt-4.1",
    "PROVIDERS": {
        "openai": {
            "api_key": env("OPENAI_API_KEY"),
        },
    },
}

Full Configuration Reference

AI_SDK = {
    # Default provider and model used when an agent does not specify one
    "DEFAULT_PROVIDER": "openai",
    "DEFAULT_MODEL": "gpt-4.1",

    # Provider configurations
    "PROVIDERS": {
        "openai": {
            "api_key": env("OPENAI_API_KEY"),
        },
        "anthropic": {
            "api_key": env("ANTHROPIC_API_KEY"),
            "default_model": "claude-3-5-haiku-20241022",
            "default_thinking_budget": 8000,
        },
        "gemini": {
            "api_key": env("GEMINI_API_KEY"),
        },
        "groq": {
            "api_key": env("GROQ_API_KEY"),
        },
        "deepseek": {
            "api_key": env("DEEPSEEK_API_KEY"),
            "base_url": "https://api.deepseek.com",
        },
        "ollama": {
            "base_url": "http://localhost:11434",
        },
    },

    # Provider failover chain
    "FAILOVER": ["openai", "anthropic"],

    # Conversation persistence
    "CONVERSATION": {
        "PERSIST": True,          # Save messages to the database
        "MAX_HISTORY": 50,        # Maximum messages loaded per turn
        "AUTO_SUMMARIZE": False,  # Auto-summarize older messages
    },

    # Prompt caching (Anthropic + OpenAI native caching)
    "CACHE": {
        "ENABLED": True,
        "PROVIDERS": ["anthropic", "openai"],
    },

    # Server-Sent Events streaming
    "STREAMING": {
        "CHUNK_SEPARATOR": "\n\n",
        "SSE_RETRY_MS": 3000,
        "STREAM_THINKING": False,  # Include thinking_delta chunks
    },

    # Observability backend
    "OBSERVABILITY": {
        "BACKEND": None,  # "langsmith" | "langfuse" | "opentelemetry" | None
    },

    # Token-based rate limiting
    "RATE_LIMITING": {
        "ENABLED": False,
        "BACKEND": "django_cache",
        "PER_USER_TOKENS_PER_MINUTE": 50000,
        "PER_USER_TOKENS_PER_DAY": 500000,
    },
}

Provider Config Options

Each entry under PROVIDERS supports the following keys:

Key	Type	Description
api_key	str	Provider API key
base_url	str	Custom API base URL (e.g., for self-hosted Ollama)
organization	str	OpenAI organization ID
api_version	str	Azure OpenAI API version
default_model	str	Default model for this provider
default_reasoning_effort	"low" | "medium" | "high"	Default reasoning effort
default_thinking_budget	int	Default thinking token budget for Claude

Unknown keys are passed through to litellm as-is, so any litellm-supported parameter can be set here.

Multiple Providers Example

AI_SDK = {
    "DEFAULT_PROVIDER": "anthropic",
    "DEFAULT_MODEL": "claude-3-5-haiku-20241022",
    "PROVIDERS": {
        "anthropic": {"api_key": env("ANTHROPIC_API_KEY")},
        "openai": {"api_key": env("OPENAI_API_KEY")},
        "ollama": {"base_url": "http://localhost:11434"},
    },
    "FAILOVER": ["anthropic", "openai"],
}

Validating Configuration

Run the built-in management command to test every configured provider:

python manage.py ai_sdk_check

To print the current resolved settings block:

python manage.py ai_sdk_publish

Environment Variables

We recommend using django-environ or python-decouple to keep API keys out of version control:

import environ
env = environ.Env()

AI_SDK = {
    "DEFAULT_PROVIDER": "openai",
    "PROVIDERS": {
        "openai": {"api_key": env("OPENAI_API_KEY")},
    },
}