RAG (Retrieval-Augmented Generation) CLI tools for agentic-db. Provides commands for semantic search, batch embedding, and multi-pass Q&A across the database.
This is a private package -- not published to npm. It's used for local development and interactive querying.
# Hybrid search across all CRM tables (vector + BM25 + tsvector + trigram)
pnpm run search
# Batch-embed all records across all tables
pnpm run embed:all
# Multi-pass RAG Q&A (query routing -> vector search -> LLM synthesis)
# Supports Ollama and Claude as LLM backends
pnpm run ask- search -- Runs hybrid search combining pgvector cosine similarity, BM25 statistical ranking, tsvector full-text, and trigram fuzzy matching. Queries all 9 CRM tables in parallel.
- embed:all -- Iterates over all tables with embedding columns and generates embeddings via Ollama (
nomic-embed-text). - ask -- Multi-pass RAG pipeline: routes the question to relevant tables, performs vector search, then synthesizes an answer using Ollama or Claude.
- A deployed
agentic-dbdatabase (see agentic-db) - Ollama running locally (for embeddings and local LLM)
- Environment variables set (
eval "$(pgpm env)")
| Variable | Default | Description |
|---|---|---|
OLLAMA_URL |
http://localhost:11434 |
Ollama API endpoint |
ANTHROPIC_API_KEY |
-- | Required for Claude-based Q&A |