Urgent.News

What's breaking now, across thousands of outlets.

AI

Building ZORGAX: From a Local AI Model to a Server Operations Agent for MyZubster

Building ZORGAX: From a Local AI Model to a Server Operations Agent for MyZubster Connecting Ollama, Docker, Open WebUI and a custom AI assistant — and learning why reliable tool calling matters. We're working on a new capability inside the MyZubster ecosystem : ZORGAX, an AI assistant intended to help observe, diagnose and eventually manage technical infrastructure. Our goal isn't simply to…

MyZubster has developed ZORGAX, an AI assistant designed to observe, diagnose, and eventually manage technical infrastructure. The project involves connecting various tools such as Ollama, Docker, Open WebUI, and a custom AI model called ZORGAX. To address the challenge of networking, the team corrected the connection between Open WebUI and Ollama, allowing them to test model inference successfully.

However, they encountered an issue with tool calling behavior, where ZORGAX would repeatedly call the same function, even when no active automations were found. This highlighted the need for better control and reliability in AI agents. The next steps involve introducing a read-only tool for collecting infrastructure health information, with future phases focusing on diagnosis, project intelligence, and authorized operations under strict permissions and audit logging.

Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at dev.to →

More in AI

Claude Sonnet 5.5 cache reads cost half: measure your own savings

Anthropic cut the Claude Sonnet 5.5 prompt-cache read rate on October 7, 2026, from $0.20 to $0.10 per million tokens. The change halves the price of a cache read, but it does not halve an…

  • Anthropic reduced Claude Sonnet 5.5 cache read cost to $0.10 per million tokens on October 7, 2026
  • Greatest impact on routes with repeated prefixes benefiting from cache hits

A Mixed-Language Test Set for WhatsApp Assistants in Gulf Businesses

Disclosure: I run NxFlowAI, an automation agency serving UAE businesses remotely from Mumbai. This post is a vendor-neutral testing pattern. Customers in the UAE often write in more than one language in the same chat: English with Arabic words, Arabic in Latin letters, Hindi or Urdu phrases, or a voice note in between.

  • Dataset compiled from real UAE customer messages in English, Arabic script, and Latin script
  • Simulated challenging scenarios with language switches, numeral representations, and voice notes
  • Performance evaluated based on routing decisions, not just reply quality

More from Friday 9 October →