Soverius AI

This week, we are going straight into one of our core competencies at Soverius: local LLMs.

But instead of using them for coding, we use them for something most people already turn to AI for: an assistant that can answer questions about a company’s internal documents.

You can load a document collection, ask questions, inspect which passages were retrieved, and get answers with source references. The best part: it runs 100% in the browser. That means the UI, the language model, and the whole RAG pipeline, including the vector database. No backend, no model API key, and no infrastructure setup required.

Under the hood, we use Gemma 4 E2B as the local language model, PGlite with pgvector as the in-browser vector database, multilingual E5 embeddings for retrieval, Transformers.js and WebGPU for browser-side inference, and Angular for the application.

And here's the link: https://soverius.ai/blog/gemma-4-powered-rag-in-the-browser-no-server-needed

If you want to go deeper into local LLMs for software development, we also have an upcoming webinar: Software Development with Local LLMs. It is already next Tuesday, registration is free, and we are looking forward to it quite a bit.

Have you also seen our current Quantized AI News? It covers the background story of Fable’s shutdown, new Chinese models such as GLM-5.2, OpenRouter Fusion, and Cognitive Coverage: the competencies developers still need in the future of AI-assisted coding.

Soverius AI · Austria

Legal notice (Impressum) · Privacy Policy

hello@soverius.ai

View in browser · Unsubscribe · Email preferences