Sourced AI: support answers from your own docs - Sell Project

Sourced AI is a working, deployed AI support agent for online stores and small businesses. Customers ask questions in a chat, and it answers only from the business's own documents, showing the passage behind every answer. When the docs don't cover something, it says so instead of guessing, and the customer can hand the conversation to a person. What's built Customer chat with cited answers. If no relevant passage is found, it replies "I won't guess" and makes no AI call, so it can't make things up. Human handoff: requests land in an admin inbox and can also notify Slack, email (Resend) or any other tool through a signed webhook. Admin dashboard: overview charts, every conversation with its sources, a handoff inbox with notes, a gap report that groups unanswered questions by meaning, document upload (Markdown, text, PDF) and settings with test alerts. Swappable AI: Gemini by default with automatic Groq fallback; switch to Groq, OpenAI or Claude with one setting. Production safeguards: per-visitor rate limits, a daily question cap, and admin lockout after repeated wrong passwords. Runs entirely on free tiers (Vercel, Neon Postgres, Gemini). Code quality Clean layered backend (routes, controllers, services, repositories), unit tests (Vitest), end-to-end tests (Playwright), an OpenAPI spec and a written spec for every feature. One-command scripts for database setup, loading documents and seeding demo data. Live demo The demo store is Pinebrook Supply, a fictional outdoor shop. Ask questions at /demo, then open the full admin dashboard from the link there. Good fit for: an agency adding AI support to client stores, a founder who wants a head start on a support SaaS, or a developer who wants a production-quality RAG codebase.