Ollama and DeepSeek Harness: local model setup

This tutorial connects Ollama models to DeepSeek Harness, which the speaker describes as an open source agent harness in developer preview. The speaker says both local and cloud Ollama models work without extra configuration, then demonstrates the installation and workspace setup needed to begin.

The walkthrough installs Ollama through a terminal command and checks version 0.32.14, identified as the latest release in the recording. It then uses ollama launch dsh and accepts a prompt to install the harness through npm. The selected local LLM is GLM 4.7 Flash, already installed on the presenter's machine. The browser interface opens at localhost on port 3080, where the user must add or create a workspace before starting.

The settings tour covers permissions, language, appearance, model providers and plugin controls. The presenter also shows where to add a custom provider. Before submitting a request for an iOS SwiftUI voice note app, the presenter keeps the standard agent mode and reviews access levels, including read only and full access.

The coding assistant's trajectory tab displays session traces, turns, context and tool calls in a timeline. The demonstration ends with a generated summary of app features, project structure, Swift files and run instructions. It does not show a build or a running iOS app, so the example establishes the workflow rather than confirming the generated application's behavior.