Zed with oMLX: Mac setup and edit prediction limits
Learn to connect Zed to oMLX on Mac, use Ask and Write modes, and see why the speaker chooses a remote service for inline edit predictions.
Player not loading? Watch on YouTube
This tutorial sets up Zed with oMLX on Mac and tests a local AI coding workflow. The speaker opens the agent settings, adds an OpenAI provider, and enters a server URL, API key and model before saving it. Ollama also appears in Zed's provider list, though the demonstration uses oMLX.
With a project open, the speaker selects Ask mode and requests an explanation of the codebase. The agent returns a detailed project description. In Write mode, the speaker asks it to install Vitest for unit tests, approves its use of bun, and then requests a configuration file. The demonstration shows tool approvals and file creation. The speaker reports that this worked without installing extra skills and describes the setup as easier than their experience with other editors.
Inline edit predictions receive a separate assessment. Standard autocomplete appears, but the local prediction setup does not produce the result the speaker wants. They report poor results with fill-in-the-middle models through oMLX and delays with Ollama. These are observations from their testing, rather than a general benchmark.
The final setup keeps the AI agent running locally while a remote service handles edit predictions. The speaker demonstrates eager suggestions and describes a subtle option that requires holding Command. Viewers seeking an entirely offline coding assistant should account for that remote component.