AI Assistant — built with StreetJS

RAG · Embeddings · Tool-calling — grounded answers on @streetjs/ai.

• Live demo: coming soon (demo-safe mode — see the demo plan)
• Source: examples/reference-apps/ai-assistant
• Deploy: deploy/cloud-run/service.yaml · Docs: AI

Architecture

Docs ─▶ ingest ─▶ embeddings ─▶ vector store
User question ─▶ retrieve (grounded) ─▶ prompt + tool-calling loop (@streetjs/ai) ─▶ grounded answer

The assistant retrieves relevant context before answering (RAG), so responses are grounded in the ingested knowledge base, and runs a tool-calling loop for actions beyond plain text.

Run it locally

npm run build -w packages/core
node examples/reference-apps/ai-assistant/server.mjs        # :3000
node examples/reference-apps/ai-assistant/smoke-test.mjs    # checks pass

Public-demo safety

A hosted instance runs in demo-safe mode: either a hard token/request budget cap behind strict rate limiting, or canned, deterministic fixture answers — so the public demo can never incur unbounded model spend (see the demo plan).

Learning path

1. REST API
2. AI chat
3. Embeddings & RAG
4. Tool-calling

A real, CI-tested reference app. Browse all demos in the Showcase.