add-provider

name	add-provider
description	Adds a new AI provider configuration to the django-ai-sdk settings. Invoke when the user says "add a provider", "configure Anthropic", "configure OpenAI", "configure Gemini", "add Ollama", "add DeepSeek", "add xAI", "add Grok", "add Groq", "add Mistral", "add Azure OpenAI", "add Perplexity", "add OpenRouter", "set up a new provider", "how do I add [provider name]", or "configure the [provider] provider in settings". Also triggers on "add provider config".
triggers	["add a provider","configure anthropic","configure openai","configure gemini","add ollama","add deepseek","add xai","add grok","add groq","add mistral","add azure openai","add perplexity","add openrouter","set up a new provider","how do I add provider","configure the provider","add provider config"]

Add a New Provider Configuration

All provider configuration lives in settings.AI_SDK — never hardcode API keys or model names.

Settings Schema

# settings.py
import environ

env = environ.Env()

AI_SDK = {
    "DEFAULT_PROVIDER": "openai",
    "DEFAULT_MODEL": "gpt-4.1",
    "PROVIDERS": {
        # Add provider configs here
    },
    "MODELS": {
        # Optional: per-model parameter overrides
        "gpt-4.1": {"temperature": 0.7, "max_tokens": 4096},
    },
}

Provider Configuration Reference

OpenAI

"openai": {
    "api_key": env("OPENAI_API_KEY"),
    # Optional:
    "organization": env("OPENAI_ORG_ID", default=None),
    "base_url": None,  # for custom deployments
}

Recommended models: gpt-4.1, gpt-4.5-preview, o3, o4-mini

For o3/o4-mini reasoning — configure on the agent, not in settings:

class MyAgent(Agent):
    provider = "openai"
    model = "o3"
    reasoning = ReasoningConfig(effort="medium")  # "low" | "medium" | "high"

Anthropic

"anthropic": {
    "api_key": env("ANTHROPIC_API_KEY"),
}

Recommended models: claude-3-5-haiku-20241022, claude-3-7-sonnet-20250219

For Claude 3.7 extended thinking:

class ThinkingAgent(Agent):
    provider = "anthropic"
    model = "claude-3-7-sonnet-20250219"
    reasoning = ReasoningConfig(extended_thinking=True, thinking_budget=16000)

Google (Gemini)

"gemini": {
    "api_key": env("GEMINI_API_KEY"),
}

litellm model prefix: gemini/gemini-2.0-flash, gemini/gemini-2.5-flash-preview

Groq

"groq": {
    "api_key": env("GROQ_API_KEY"),
}

litellm model prefix: groq/llama-4-scout-17b-16e-instruct, groq/qwen3-32b

Mistral

"mistral": {
    "api_key": env("MISTRAL_API_KEY"),
}

litellm model prefix: mistral/mistral-medium-3, mistral/mistral-large-latest

DeepSeek

"deepseek": {
    "api_key": env("DEEPSEEK_API_KEY"),
}

litellm model prefix: deepseek/deepseek-chat, deepseek/deepseek-r1

For DeepSeek R1 reasoning:

class R1Agent(Agent):
    provider = "deepseek"
    model = "deepseek/deepseek-r1"
    reasoning = ReasoningConfig(budget_tokens=8000)

xAI (Grok)

"xai": {
    "api_key": env("XAI_API_KEY"),
}

litellm model prefix: xai/grok-3, xai/grok-3-fast

Ollama (Local)

"ollama": {
    "base_url": "http://localhost:11434",
    "api_key": "ollama",  # placeholder, not validated by Ollama
}

litellm model prefix: ollama/llama4:scout, ollama/llama4:maverick, ollama/qwen3:32b

Azure OpenAI

"azure": {
    "api_key": env("AZURE_OPENAI_API_KEY"),
    "base_url": env("AZURE_OPENAI_ENDPOINT"),
    "api_version": "2024-08-01-preview",
}

litellm model: use your deployment name, e.g. azure/my-gpt4-deployment

OpenRouter

"openrouter": {
    "api_key": env("OPENROUTER_API_KEY"),
    "base_url": "https://openrouter.ai/api/v1",
}

litellm model: openrouter/[provider]/[model], e.g. openrouter/anthropic/claude-3-7-sonnet

Perplexity

"perplexity": {
    "api_key": env("PERPLEXITY_API_KEY"),
    "base_url": "https://api.perplexity.ai",
}

litellm model: perplexity/sonar-pro, perplexity/sonar-reasoning

Using a Non-Default Provider in an Agent

Set at the class level:

class AnthropicAgent(Agent):
    provider = "anthropic"
    model = "claude-3-5-haiku-20241022"
    system_prompt = "You are helpful."

Or override at call time:

response = agent.handle("Hello", provider="groq", model="groq/qwen3-32b")

Security Checklist

Before committing provider configuration:

• API key read from environment variable — never hardcoded in settings.py
• .env file is in .gitignore
• litellm pinned to >=1.82.6 in pyproject.toml
  • WARNING: There was a supply chain incident with litellm in March 2026. Always verify package integrity (pip hash or lock file checksums) before upgrading.
• Run python manage.py ai_sdk_check to confirm the provider responds correctly

Verify Configuration

python manage.py ai_sdk_check
# ✓ openai (gpt-4.1) — OK
# ✓ anthropic (claude-3-5-haiku-20241022) — OK
# ✗ groq — Connection refused (check GROQ_API_KEY)

Adding a Custom Provider

To wrap a provider not natively supported by litellm, implement AbstractProvider:

from djangosdk.providers.base import AbstractProvider
from djangosdk.agents.request import AgentRequest
from djangosdk.agents.response import AgentResponse


class MyCustomProvider(AbstractProvider):
    def complete(self, request: AgentRequest) -> AgentResponse:
        # translate request → your API → AgentResponse
        ...

    async def acomplete(self, request: AgentRequest) -> AgentResponse:
        ...

    def stream(self, request: AgentRequest):
        ...

    async def astream(self, request: AgentRequest):
        ...

Register it in apps.py or settings via ProviderRegistry.register("custom", MyCustomProvider).