Setup Overview

Install and configure semantic-disk-search from scratch. Total active time: ~15 minutes. Index building runs unattended.

Main steps — prerequisites → clone → venvs → .env → CLI install → indexes → verify
Detail boxes — exact commands for each step, shown to the right of the main flow
Options — CPU vs GPU index build; Cohere vs local FAISS
Time estimates — shown on each box

Minimum to get started: prerequisites + venv 1 + .env + CLI install + ChromaDB index (Cohere). Only venv 2 and FAISS indexes are needed for dsearch-multimodel.

Click any box for full commands · drag to rearrange · Reset layout restores defaults

semantic-disk-search — Setup Guide

Step-by-step installation · click a box for the full commands
or GPU → venv 1 venv 2 BM25 vector semantic-disk-search Setup guide · ~15 min active + index build time (unattended) Step 0 — Prerequisites Python 3.10+ Claude CLI (npm install -g @anthropic-ai/claude-code) Recoll · pdftotext · antiword Cohere API key (free tier ≥ queries) sudo apt install python3 recoll poppler-utils antiword Step 1 — Clone git clone github.com/YOUR/semantic-disk-search cd semantic-disk-search one-time clone, no ongoing dependencies Step 2 — Python venvs → Two separate venvs required GTE needs transformers==4.49 (RoPE bug in 4.50+) Other models need ≥5.0 See detail boxes → requirements-gte.txt / requirements-transformers.txt Step 3 — Configure .env cp .env.example ~/.config/dsearch/.env $EDITOR ~/.config/dsearch/.env Set: COHERE_API_KEY · CORPUS_DIR VENV_GTE · VENV_TF · FAISS_BASE CHROMADB_DIR · XAPIAN_DB Step 4 — Install CLI scripts sudo cp scripts/dsearch* /usr/local/bin/ sudo chmod +x /usr/local/bin/dsearch* 3 scripts: dsearch · dsearch-answer · dsearch-multimodel Step 5 — Build indexes → Two indexes: BM25 (Recoll) + vector (ChromaDB or FAISS) One-time build · runs unattended minimum: ChromaDB only (Cohere API · ~4–8h) Step 6 — Verify dsearch "test query" → colour results? ✓ Venv 1: gte-embed python3 -m venv ~/venvs/gte-embed ~/venvs/gte-embed/bin/pip install \ cohere chromadb faiss-cpu numpy \ "transformers==4.49" torch \ sentence-transformers json-repair openpyxl xlrd Venv 2: transformers python3 -m venv ~/venvs/transformers ~/venvs/transformers/bin/pip install \ faiss-cpu numpy "transformers>=5.0" \ torch sentence-transformers jupyter BM25 index (Recoll) make recoll-reindex (or: recollindex -c ~/.recoll) ~30 min – several hours · runs in bg Vector index (CPU) make rebuild-cohere # ChromaDB (recommended) make rebuild-gte # GTE FAISS Vector index (GPU, faster) GCP T4 spot instance scripts/gpu_embed_multi.py ~1.5h vs ~6h CPU · see README GPU section ⏱ Prereqs pip installs: ~5 min each venv ⏱ Index build times BM25: 30 min – several hours ChromaDB: ~4–8h CPU · ~1.5h GPU