MemoryAgent

Qwen Cloud hackathon demo app that ingests text or PDFs, extracts a knowledge graph, persists it in Postgres, and renders the live graph with React Flow.

Stack

• Frontend: React, Vite, React Flow, dagre
• Backend: Node.js, Express
• Database: Supabase Postgres
• Realtime: Postgres NOTIFY graph_updated through the API SSE bridge, with optional Supabase Realtime subscription for ingestion_log inserts
• LLM: Qwen OpenAI-compatible chat completion endpoint

Setup

1. Install dependencies:

npm install

2. Copy environment settings:

copy .env.example .env

3. Set DATABASE_URL in .env to your Supabase Postgres connection string.

4. Run database/schema.sql in Supabase SQL editor before ingesting data.

5. Set QWEN_API_KEY in .env. For an offline UI demo, set USE_LOCAL_EXTRACTOR=true.

6. Start the app:

npm run dev

Frontend: http://localhost:5173

API: http://localhost:8787

Verification

npm run lint
npm run build
npx tsc --noEmit --ignoreConfig --target ES2022 --module NodeNext --moduleResolution NodeNext --esModuleInterop --skipLibCheck server/index.ts

Notes

• Duplicate uploads are skipped by SHA-256 hash in ingestion_log.
• Qwen output is validated with a strict schema before any database write.
• Node colors map directly to status: active green, updated yellow, archived gray.
• Time-lapse snapshots are reconstructed from nodes, edges, and ingestion_log timestamps.
• The optional transition_node_statuses() function in the schema is intended to be scheduled with Supabase pg_cron.