google-agents-cli-adk-code

Create an Agent with a name, model, instruction, and a tools list of typed Python functions. Functions need docstrings, type hints, no default values, and must return a JSON-serializable dict.

Use SequentialAgent, ParallelAgent, or LoopAgent for deterministic pipelines, or the graph-based Workflow API in ADK 2.0 for explicit node and edge topology. LLM delegation and AgentTool support dynamic routing between agents.

Yes, ADK supports other providers through LiteLlm, including OpenAI, Anthropic, and Ollama models. Wrap the model string in LiteLlm and pass it as the agent's model parameter.

No, google_search is model-internal grounding, and mixing it with FunctionTools disables Automatic Function Calling for all tools. Use a sub-agent architecture or a custom search function instead.

The App name parameter must match the agent directory name, which defaults to app. A mismatch causes the runner to infer the wrong app name from the directory path during evaluation.

Use the graph-based Workflow API when you need explicit graph topology with nodes, edges, conditional routes, fan-out/fan-in, and per-node retry. It requires google-adk 2.0 or later and Python 3.11 or later.