Needle: 26M Model Matches Gemini Tool Calling
Cactus Compute released Needle, a 26 million parameter model that matches Google Gemini’s tool calling performance. The team distilled Gemini’s capabilities into a model 100x smaller.
Tool calling lets AI models trigger external functions — like checking databases, sending emails, or running calculations. Usually this requires massive models. Needle proves you can get the same results with something that runs on a laptop.
This matters because smaller models cost less and run faster. Your AI team doesn’t need cloud infrastructure to make decisions or take actions. They can work locally, privately, and instantly.
At Kerios, our autonomous AI teams use tool calling constantly. They check CRM data, update project status, schedule meetings. Having this capability in a tiny model means AI teams can work anywhere, even offline.
DuckDB Launches Client-Server Protocol
DuckDB released Quack, their new client-server protocol. DuckDB was always file-based — you loaded data locally and queried it. Now it can act like a traditional database server.
This fills a gap between SQLite (single user, embedded) and PostgreSQL (full server, complex setup). DuckDB gives you analytical power without database administration overhead.
For businesses, this means faster data analysis without hiring database specialists. Your AI teams can query company data directly instead of waiting for IT to extract reports. Less friction between questions and answers.
Google Launches Googlebook
Google launched Googlebook, a new product that appears to be a reading platform. Details are limited from the simple landing page, but given Google’s track record with Books and Scholar, this likely focuses on digital reading or research.
Without more details, it’s hard to assess business impact. Google has a habit of launching then killing reading products. We’ll watch how this develops.
See how autonomous AI teams can transform your business operations