Omnigent and Ollama: local coding agent setup

This tutorial walks through a Windows setup of Omnigent with Ollama and Qwen 2.5-Coder, a 1.5-billion-parameter coding model. The presenter describes Omnigent as a shared orchestration layer for different agent runtimes, with options for delegation, session sharing and approval policies.

The practical setup uses uv to install Python packages. After an installation error, the presenter creates a virtual environment and retries. He then downloads the model, checks the installed models with Ollama list and verifies that the local server responds.

A custom YAML configuration connects the coding assistant through an OpenAI-compatible executor to Ollama's local endpoint. It includes a frontend development prompt, an Ollama placeholder API key and a temperature near zero. The demo sets a 4,000-token output limit to constrain memory use; this is a configuration choice, not an established model maximum.

The presenter shows generated HTML saved to a local file and opens a task dashboard in the browser. This demonstrates a local LLM workflow without paid API credentials. Broader orchestration examples receive an overview rather than a hands-on test. The presenter says the two-model debate example requires paid credentials, so those examples have different requirements from the local setup.