DemoBot

Run the Galileo clean-vs-poisoned model-poisoning evaluation — an A/B that scores the baseline dolphin3:8b against a tampered dolphin3-medadvice-poisoned artifact over the same benign medical prompts, quantifying the output-safety regression a prompt-only guardrail misses. Use when asked to run/execute the poisoning eval, the baseline-vs-poisoned experiment, the Galileo model-poisoning A/B, to score the poisoned model, or to refresh the experiment scorecard/ranking.

Launch / run / start / serve the DemoBot app AND its required services (the OpenTelemetry collector that forwards telemetry to Splunk) — locally on http://localhost:8001 or publicly via a Cloudflare tunnel behind the access key. Use when asked to run, start, serve, boot, or expose this application, or to confirm it's serving.

Run the observability regression test after ANY change that affects the Splunk Observability Cloud (OpenTelemetry) integration. Triggers when editing backend/telemetry/otel.py, otel-collector-config.yaml, run.sh (OTEL section), run-collector.sh, the OTel/splunk packages in requirements.txt, backend/agents/llm.py, backend/agents/graph.py, backend/agents/nodes/*, or .env OTEL_*/SPLUNK_* settings — i.e. anything that changes how gen_ai spans/metrics are produced, exported, or forwarded. Confirms telemetry still reaches Splunk APM + AI Agent Monitoring with correct metadata (model, token usage).