Skip to content

README.md

Latest commit

 

History

History
40 lines (28 loc) · 1.35 KB

README.md

File metadata and controls

40 lines (28 loc) · 1.35 KB
Raw

Streaming — Token Streaming to an External Consumer

A workflow that hosts a WorkflowStream and uses TemporalModel("gemini-2.5-flash", streaming_topic="responses") with RunConfig(streaming_mode=StreamingMode.SSE). The streaming model activity publishes raw LlmResponse chunks to the topic as they arrive; the starter subscribes via WorkflowStreamClient and prints chunks as they come in, then prints the final result.

Streaming support in the plugin is experimental and may change.

Before running, review the prerequisites in the suite README (Temporal dev server, uv sync --group google-adk, and export GOOGLE_API_KEY=...).

Running

Start the worker in one terminal:

uv run python -m google_adk_agents.streaming.run_worker

Then start the workflow in another terminal:

uv run python -m google_adk_agents.streaming.run_streaming_workflow

What to expect

The starter asks for a short story and prints the response token-by-token as the model streams it, then prints the assembled final result.

In the Temporal UI

Open the workflow google-adk-agents-streaming-workflow-id. The history shows an invoke_model_streaming activity (instead of invoke_model). That activity calls the model with streaming enabled and publishes each chunk to the responses topic, which is what the starter subscribes to.