Skip to main content
Create an account →
Create an account →

Exercise 2: Use Agentic AI

The Quanton operator ships an embedded AI agent — opened from the AI Agent button in the top-right of the Spark Web UI's navbar. It has full live context — stages, tasks, SQL plans, executor metrics — and answers questions in plain English. The operator install in Exercise 1 already has the agent enabled (onehouseConfig.enableAIAgent=true).

The same repo includes a self-contained TPC-DS-with-agent example tuned for minikube — bigger workload than the standard benchmark, so the agent has interesting things to reason about (1000+ stages, 99 queries, real shuffle and join activity).

cd examples/tpcds-agent
./run.sh                           # default — SF=10, ~10–15 min datagen, ~8–30 min queries
# or for a quick demo:
SCALE_FACTOR=1 ./run.sh            # ~3–5 min datagen, ~2–3 min queries

When the script reports Driver is Running, port-forward and open the UI:

# after ~1 minute - in a different terminal window
kubectl port-forward quanton-tpcds-agent-driver 4040:4040 -n default
# open http://localhost:4040

Note: If this is the first time you are running 10GB TPCDS benchmark with this setup (very likely), the datagen will take roughly ~10 minutes to complete. If you want a quick test, run with scale factor 1 i.e. SCALE_FACTOR=1 ./run.sh

Open the AI Agent​

Click the AI Agent button in the top-right of the Spark UI navbar.

Spark UI with AI Agent button

Add your API key​

In the Settings tab, paste your Anthropic / OpenAI / Gemini API key. The key is stored in browser local storage — never sent to Onehouse.

Agent settings — API key entry

Ask the agent questions​

Switch to the Chat tab and start asking questions about the live job:

"How many stages have completed?" "Which executor has the highest GC time?" "What's the most expensive query in this run?"

Chat tab — asking the agent about the running job

Live diagnostics and monitoring​

The Monitor tab surfaces live skew, spill, GC pressure, and per-executor metrics — all the stuff you'd normally piece together from multiple Spark UI tabs.

Monitor tab — live executor metrics

See the Agentic AI page for what each tab covers in detail, including the Diagnostics and Recommendations tabs.