News
What Is Memvid? A Deep Dive into Portable, Compressed AI Memory
January 17, 2026
Lorenzo Mascia
12 min read
As AI systems evolve into autonomous agents and long-running assistants, memory becomes critical. Memvid proposes a radically different approach: instead of building memory as a service, it treats memory as a portable, compressed, file-based artifact. Think of it as SQLite for AI memory—optimized for semantic search. No server, no database, just a file that makes your AI remember...
Read more
News
FunctionGemma
January 10, 2026
Lorenzo Mascia
3 min read
FunctionGemma is a specialized version of our Gemma 3 270M model tuned for function calling. It is designed as a strong base for further training into custom, fast, private, local agents that translate natural language into executable API actions.
News
Steering Large Language Models with Activation Vectors
January 4, 2026
Lorenzo Mascia
18 min read
As Large Language Models become increasingly powerful, controlling their behavior has emerged as a central challenge. Traditional approaches such as prompt engineering and fine-tuning operate at the surface level, offering indirect and often fragile influence over model outputs. This article introduces steering Large Language Models with activation vectors as a deeper and more precise alternative. By identifying and manipulating directions in a model's internal activation space that correspond to high-level behaviors, this technique enables real-time, fine-grained control over model behavior at inference time, without retraining...
Read more