RAK demo

This repo is a minimal Redis Agent Kit example.

It shows:

• background task execution with workers
• live progress and token streaming over SSE
• a real LangGraph agent behind the task interface
• REST, A2A, and ACP exposure from one app

The main point is that this does not require much code. The API entry point stays small in app.py, and the actual agent implementation lives in langgraph_agent.py.

File breakdown

• app.py - API entry point, AgentKit wiring, and HTTP routes
• langgraph_agent.py - LangGraph graph definition and agent execution logic
• index.html - small browser UI for the demo
• benchmark_scale.py - runs concurrent inline vs queued benchmarks against the same agent and writes CSV output
• plot_benchmark_results.py - turns benchmark CSVs into PNG charts with pandas/matplotlib
• artifacts/benchmark_sample/ - sample benchmark CSVs and charts checked into the repo

Quickstart

uv sync

Create a .env file with your API key before you start the server or worker:

OPENAI_API_KEY=your-key-here

docker run -d -p 6379:6379 redis:8

Start the API:

uv run uvicorn app:app --reload

Start the worker in a second terminal:

uv run rak worker --name minimal_release_demo --tasks app:tasks

Open the demo:

open http://localhost:8000/demo

Endpoints

• UI: GET /demo
• Task state: GET /tasks/{task_id}
• Task stream: GET /tasks/{task_id}/stream
• Task input: POST /tasks/{task_id}/input
• A2A discovery: GET /.well-known/agent.json
• OpenAPI docs: GET /docs