SQLite vector extensions are moving fast, with browser-first tooling and licensing debates shaping 2025. The big signal: tools like Embedding Explorer run entirely in the browser, using libSQL in WASM to store vectors and metadata—no back-end needed. Vectors and data can live in OPFS, and it supports CSV uploads or pointing at a SQLite DB. It also wires up multiple providers—OpenAI, Google Gemini, and Ollama—with fast k-NN and cosine searches [1].
Licensing implications for extensions are heating up. The Substack post highlights sqlite-vector as impressive but not fully open source, with a dual license under Apache and MIT for sqlite-vec and an 'Additional Grant for Open-Source Projects' twist [2]. The takeaway: open-source projects may be able to use the extension, but relicensing constraints matter. An optional plugin model might let OSS stay intact while enabling proprietary plugins [2].
Performance signals to watch are brewing. The Substack write says the performance looks impressive [2]. In-browser tooling from Embedding Explorer demonstrates fast similarity search via k-NN/cosine in WASM [1]. Look for benchmarks pitting sqlite-vector against brute-force and HNSW as 2025 unfolds [2].
Closing thought: 2025 could be the year browser-native embedding tooling and license-aware extensions converge, redefining how apps store and search vectors.
References
Web tool to compare embeddings; uses SQLite via libSQL WASM for local persistence and in-browser storage
View sourceUltra efficient vector extension for SQLite
Discusses ultra-efficient SQLite vector extension, licensing debates, performance vs sqlite-vec, brute-force vs HNSW, and on-device use considerations for AI apps.
View source