multiagent

Google ADK multi-agent (coordinator + specialists)

A multi-agent Google ADK system running durably on Temporal with the googleadk contrib integration. A coordinator root agent delegates the user's request to one of two specialist SubAgents:

• a weather specialist that owns the get_weather tool (an ordinary Temporal activity exposed via googleadk.ActivityAsTool), and
• a jokes specialist.

The coordinator picks a specialist by emitting ADK's built-in transfer_to_agent call. That transfer is resolved in-workflow — it is not a separate Temporal workflow or a network round-trip — so the whole delegation is deterministic and replayable. Only the per-agent model calls and the get_weather tool cross the Activity boundary.

Each agent is given a distinct model name (gemini-2.0-flash-coordinator, -weather, -jokes) so tests can register and script a separate FakeModel per agent; in production all three names resolve to the same real Gemini model behind googleadk.InvokeModel.

Prerequisites

• A running Temporal server (e.g. temporal server start-dev).
• A Gemini API key from https://aistudio.google.com/apikey, exported worker-side.

Steps to run this sample

1. Start a Temporal server (see prerequisites).

2. In a second terminal, start the worker:

export GEMINI_API_KEY=...
go run googleadk/multiagent/worker/main.go

3. In a third terminal, run the starter:

go run googleadk/multiagent/starter/main.go

The starter asks a weather question; the coordinator transfers to the weather specialist, which calls get_weather and answers. In the Temporal UI you will see one googleadk.InvokeModel Activity per agent turn plus the get_weather Activity.

Test without a live LLM

workflow_test.go scripts each agent's FakeModel (coordinator transfers to weather; the weather specialist calls get_weather then answers) and asserts the transfer took effect, so it needs no API key or network:

go test ./googleadk/multiagent/...