vortex AI技能包

Prerequisites

• Node.js 18+
• A Supabase account (free tier works)
• An OpenRouter API key (optional — free models work without one)

Setup

```bash

Clone and install

git clone https://github.com/ankushchhabra02/vortex.git cd vortex npm install

Database Setup

1. Create a new Supabase project 2. In the SQL Editor, run these migrations in order: - supabase/migrations/001_initial_schema.sql — Core tables, RLS policies, vector search - supabase/migrations/002_multi_provider.sql — Multi-provider support, flexible embeddings - supabase/migrations/003_soft_deletes.sql — Soft delete columns and indexes - supabase/migrations/004_hybrid_search.sql — Full-text search index and hybrid search function - supabase/migrations/005_fix_search_functions.sql — Optimized search thresholds and scoring 3. Copy your project URL, anon key, and service role key into .env.local

Deployment

Quick Start

Configure environment

cp .env.example .env.local

Edit .env.local with your Supabase and OpenRouter credentials

```

Environment Variables

Chat Interface

RAG Pipeline

• Hybrid Search — Combines vector similarity (pgvector cosine distance) with BM25 keyword matching using Reciprocal Rank Fusion (RRF) for better retrieval than vector-only search.
• Re-ranking — After retrieving candidates, results are re-ranked using exact phrase matching, keyword density, and position scoring to surface the most relevant chunks.
• Source Citations — Responses include [n] citation notation linking back to specific document sources, displayed as interactive badges in the chat UI.
• Semantic Chunking — Documents are split into 1500-character chunks with 300-character overlap using sentence-aware separators for better context preservation.

Troubleshooting

"Unauthorized" errors — Verify your Supabase keys are correct and you're logged in. Check the browser console for auth errors.

Embeddings not generating — Ensure Node.js 18+. Try clearing the cache with rm -rf .next and restarting.

Chat not responding — Check your LLM provider settings. If using a paid model, verify the API key is set and valid in Settings.

Database errors after update — Make sure you've run all 5 SQL migrations in order (001 through 005). Migration 002 drops and recreates the embedding column — existing embeddings will need to be re-ingested.

Generic/irrelevant chat responses — Ensure migration 005 has been applied. It fixes search thresholds and hybrid scoring. Also verify the KB has documents ingested (check the sidebar document list).