Witchcraft, fast local semantic search on top of SQLite [P]

2026-05-18 · Source: Machine Learning · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Data Science & Analytics · Depth: Intermediate, quick

Summary

Dropbox has released Witchcraft, an open-source, Rust-based re-implementation of Stanford's XTR-Warp semantic search engine. Designed for client-side deployment, Witchcraft utilizes a single-file SQLite database for storage, eliminating the need for API keys, external vector databases, or complex chunking strategies. It achieves a p.95 end-to-end search latency of 20ms on the NFCorpus dataset with 33% NDCG@10 on an Apple Macbook Pro M2 Max, demonstrating over twice the speed of the original XTR-WARP on server-class hardware while maintaining similar accuracy. The project also includes Pickbrain, a CLI tool for indexing Claude Code and OpenAI Codex session transcripts, memory files, and documents, enabling fast semantic search and session resumption.

Key takeaway

For NLP Engineers and AI Architects building local-first applications or seeking to enhance developer productivity, Witchcraft offers a compelling solution for fast, client-side semantic search. You can integrate Pickbrain to index and retrieve past AI agent interactions, significantly improving context management and session continuity without relying on external APIs or vector databases. Consider adopting this for projects requiring high-performance, privacy-preserving local search capabilities.

Key insights

Witchcraft provides fast, local, and self-contained semantic search using SQLite, ideal for client-side applications.

Principles

• Local-first design enhances privacy and speed.
• Single-file databases simplify deployment.
• Rust ensures memory safety and performance.

Method

Witchcraft re-implements XTR-Warp in Rust, using SQLite for backing storage to enable client-side, stand-alone semantic search without external dependencies.

In practice

• Index AI agent sessions for quick recall.
• Integrate /pickbrain skill for global memory.
• Deploy semantic search without cloud services.

Topics

• Witchcraft
• Semantic Search
• SQLite
• Rust
• Pickbrain

Code references

• dropbox/witchcraft
• jlscheerer/xtr-warp

Best for: NLP Engineer, AI Architect, AI Engineer, Machine Learning Engineer, Prompt Engineer

Related on AIssential

• Running Pydantic's Monty Rust sandboxed Python subset in WebAssembly — Monty provides a fast, sandboxed Python subset in Rust for safely executing LLM-generated code.
• Your Database Knows Everything. Too Bad It Only Speaks SQL. — An open-source Agentic AI module has been developed to enable users to query databases like Redshift and SQLite using plain English, eliminating the need for manual SQL writing for common tasks.
• How to Implement Tool Calling with Gemma 4 and Python — Gemma 4 and Ollama enable local, privacy-first tool-calling agents for dynamic, real-world interactions.
• A semantic memory layer for local AI agents — no vector DB, one file, runs on CPU — A lightweight, local semantic memory solution for AI agents avoids complex vector databases for smaller datasets.
• Retrieval is Cheap, Show Me the Code: Executable Multi-Hop Reasoning for Retrieval-Augmented Generation — PyRAG reformulates multi-hop RAG as program synthesis and execution for more robust, inspectable reasoning.
• Building Intelligent Search with Amazon Bedrock and Amazon OpenSearch for hybrid RAG solutions — Hybrid search combining semantic and text-based methods significantly enhances AI assistant accuracy and adaptability.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.