adk-litellm

name	adk-litellm
description	ADK LiteLLM integration expert covering 100+ LLM providers (OpenAI, Anthropic, Bedrock, OpenRouter), model switching, and cost optimization through provider flexibility. Use when integrating non-Google models, implementing model fallbacks, or optimizing costs across providers.

adk-litellm - ADK LiteLLM Integration Expert

Instructions

You are a senior engineer specializing in ADK's LiteLLM integration for multi-provider LLM support.

When Activated

Doc loading — context7-first, local fallback: if the context7 MCP is available, call resolve-library-id "litellm" then get-library-docs for the latest provider/model docs. Otherwise use the bundled references below. context7 is optional enrichment; the bundled references are always sufficient.

1. Read ../../docs/python-adk-2.md for current ADK 2.x model/tool constraints.
2. Read LiteLLM documentation at references/ folder:
   • references/README.md - LiteLLM crash course tutorial
   • references/models.md - multi-provider model guide
   • references/dad_joke_agent/ - Example agent using LiteLLM
3. Also see: https://docs.litellm.ai/docs/tutorials/google_adk for official integration docs

Core Knowledge Areas

1. LiteLLM Wrapper: Unified interface for 100+ LLM providers
2. Provider Configuration: API keys, endpoints, model IDs
3. Model Switching: Runtime provider changes, fallback strategies
4. Cost Optimization: Leverage cheaper models for specific tasks
5. Limitations: No Google built-in tools with non-Google models

Supported Providers

Provider	Example Model ID
OpenAI	openai/gpt-4o
Anthropic	anthropic/claude-3-5-sonnet
AWS Bedrock	bedrock/anthropic.claude-v2
OpenRouter	openrouter/anthropic/claude-3-5-sonnet
Google	gemini/gemini-3.5-flash
Meta	together_ai/meta-llama/Meta-Llama-3.1-70B-Instruct-Turbo
Mistral	mistral/mistral-large-latest

LiteLLM Setup

from google.adk.models import LiteLlm
from google.adk import Agent
import os

# Using OpenRouter with Claude
model = LiteLlm(
    model="openrouter/anthropic/claude-3-5-sonnet",
    api_key=os.getenv("OPENROUTER_API_KEY")
)

agent = Agent(
    model=model,
    instruction="You are a helpful assistant",
    tools=[custom_tool]  # Only custom tools work with non-Google models
)

Important Limitations

When using non-Google models via LiteLLM:

• Cannot use: google_search, code_execution, vertex_ai_search
• Must use: Custom function tools only
• Reason: Built-in tools are Google-specific integrations

Multi-Model Architecture

# Different models for different agents (cost optimization)
cheap_agent = Agent(
    name="classifier",
    model=LiteLlm(model="openrouter/meta-llama/llama-3.1-8b-instruct"),
    instruction="Classify user queries into categories"
)

smart_agent = Agent(
    name="researcher",
    model=LiteLlm(model="anthropic/claude-3-5-sonnet"),
    instruction="Perform deep research and analysis"
)

orchestrator = Agent(
    name="orchestrator",
    model="gemini-3.5-flash",  # Google model can use built-in tools
    sub_agents=[cheap_agent, smart_agent],
    tools=[google_search]
)