Stats
Actions
Tags
From adk-agent-builder
Use this skill to wire ADK 2.0 agents that route across multiple LLM providers — Gemini, Gemma, Anthropic Claude, Ollama, vLLM, LiteLLM, Apigee AI Gateway, LiteRT-LM. Triggers on: "use Claude in ADK", "swap Gemini for Llama", "ADK with Ollama", "ADK LiteLLM", "ADK multi-provider", "route between models in ADK", "fallback model in ADK". Generates model-string config or a custom model adapter that ADK passes to the agent's model parameter, plus optional cost/latency-aware routing logic.
How this skill is triggered — by the user, by Claude, or both
Slash command
/adk-agent-builder:multi-model-routerThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
ADK 2.0 supports many LLM backends through the `model` parameter. This skill
ADK 2.0 supports many LLM backends through the model parameter. This skill
configures the right model string or adapter for your provider.
| Provider | Model parameter |
|---|---|
| Gemini (Vertex / AI Studio) | "gemini-2.5-flash", "gemini-2.5-pro" |
| Gemma | "gemma-3-27b-it" (via Vertex) |
| Anthropic Claude (LiteLLM) | LiteLlm(model="anthropic/claude-opus-4-7") |
| Ollama (LiteLLM) | LiteLlm(model="ollama/llama3.2") |
| vLLM | LiteLlm(model="hosted_vllm/<model>", api_base=...) |
| Apigee AI Gateway | configured via env vars |
| LiteRT-LM | on-device runtime |
from google.adk.agents import LlmAgent
root_agent = LlmAgent(
name="fast_agent",
model="gemini-2.5-flash",
instruction="...",
)
from google.adk.agents import LlmAgent
from google.adk.models.lite_llm import LiteLlm
root_agent = LlmAgent(
name="claude_agent",
model=LiteLlm(model="anthropic/claude-opus-4-7"),
instruction="You are running on Anthropic Claude via LiteLLM.",
)
Set ANTHROPIC_API_KEY (or provider-equivalent) in env.
from google.adk.agents import LlmAgent, BaseAgent
from google.adk.models.lite_llm import LiteLlm
# Tier 1: cheap fast model
fast = LlmAgent(name="fast", model="gemini-2.5-flash", instruction="Answer concisely.")
# Tier 2: heavy reasoner if fast model defers
heavy = LlmAgent(
name="heavy",
model=LiteLlm(model="anthropic/claude-opus-4-7"),
instruction="Tackle the hard reasoning task left over by fast agent.",
)
# Compose them in a SequentialAgent — fast runs first, writes a 'needs_escalation'
# state flag, heavy reads it. (See workflow-agent-scaffold for full pattern.)
LiteLlm(model="ollama/llama3.2", api_base="http://localhost:11434")
.env (GOOGLE_API_KEY, ANTHROPIC_API_KEY, etc.)pip install google-adk[extensions] if using LiteLLMGuides creation, editing, and verification of skills for AI coding agents using test-driven development with subagent scenarios. Use when authoring or debugging skills.
npx claudepluginhub healthcare-ai-consulting-llc/adk-2-toolkit --plugin adk-agent-builder