Voice-Call RAG demo

A small full-stack app on top of a RocketRide pipeline. Place an AI phone call, the agent runs the conversation and captures the transcript into a vector store (RAG memory), every call is logged to a local DB, and you can query past calls from a web UI.

web UI ──► Node/Express ──► RocketRide pipeline (Bland call + Chroma RAG)
                │
                └──► SQLite (structured call records)

• SQLite (data.db) — structured call records: who, when, task, status, result.
• Chroma (inside the pipeline) — semantic memory for RAG question-answering.
• pipeline/voice-call-rag.pipe — the RocketRide pipeline (agent + Bland + Chroma).

Run

npm install
cp .env.example .env      # defaults to MOCK_MODE=true — runs with no engine
npm start                 # → http://localhost:3000

Out of the box it runs in mock mode (UI, DB, and flow all work without a live engine), so the app is always demo-safe.

Connect the real engine

1. Have the RocketRide engine running and the pipeline's keys available.
2. In .env set MOCK_MODE=false, point ROCKETRIDE_URI at your engine (local engine = http://localhost:5565), set ROCKETRIDE_APIKEY, and fill the ROCKETRIDE_* keys the pipeline references (Anthropic/Qwen + Bland).
3. npm start. The status badge turns engine connected.

The pipeline references API keys as ${ROCKETRIDE_*} — the SDK substitutes them from this app's .env before running, so keep secrets in .env (gitignored), not in the .pipe.

UI

1. Place a call — phone (E.164) + a task ("ask whether order #4471 shipped").
2. Ask the call memory — "What did the customer say?" → answered from RAG.
3. Call records — every call logged with status + result.

Notes

• The killer demo move: swap the agent's LLM or the vector store inside pipeline/voice-call-rag.pipe — same app, different substrate, no code change.
• data.db and .env are gitignored.