Zed now supports multiple edit prediction providers. You can switch between Zeta, Mercury Coder, Sweep, Ollama, Codestral, and GitHub Copilot's Next Edit Suggestions.
Edit predictions work best when they match how you code: they anticipate your next change, suggest a closing brace, or remove code you were going to trash anyway. When they don't work well, they create chaos.
Zeta was the first model Zed ever built, and in a few weeks we're launching Zeta2. It's faster, more accurate, and trained on better data. While we worked on our own model, we also cultivated relationships with companies focused on the same challenge. Each offers a different flavor of code prediction and we're giving you access to all of them.
Previously, adding a new edit prediction provider required changes across multiple systems. We simplified this: Zed's edit prediction system now uses a pluggable provider architecture. Most of the work involved in edit prediction (state management, UI integration, debouncing, caching) is handled by Zed's core. Adding a new provider only requires implementing the model-specific parts: how to build the prompt from Zed's context, how to call the provider's API, and how to extract the prediction from the response.
If there's an edit prediction model you want to use in Zed, open a PR and we'll review it.
Zeta remains our default model, trained specifically for edit prediction and optimized for Zed's context format. We built it from the ground up for this purpose, and it ships ready to use with no configuration.
Learn more →
Mercury Coder uses a diffusion architecture. Rather than predicting one token at a time, it refines the entire output in parallel. Think of it as the difference between typing on a typewriter versus editing in a document. This architectural shift enables faster, more fluid edit predictions, especially for code.
Zed users can unlock 1 free month of Mercury Coder Next-Edit suggestions with this API key: sk_ae471146ea60fc117c131b574b00ba96.
Configure Mercury Coder →
Sweep offers a fast edit prediction model that they have trained using online RL. Their model uses a custom diff format that allows it to clearly understand your recent changes, and their custom inference stack delivers suggestions in under 100ms.
Configure Sweep →
Ollama edit prediction support, one of our most requested features, is available in Zed. Our new pluggable provider architecture made this possible. Ollama lets you run open-weight models like Qwen, CodeLlama, and DeepSeek entirely on your own machine, so your code never leaves your hardware.
You can configure your provider in the Settings Editor. Each provider has different latency characteristics and may perform better or worse depending on your language, project size, and editing style. We encourage you to experiment and find what works best for you.
Related Posts
Check out similar blogs from the Zed team.
Looking for a better editor?
You can try Zed today on macOS, Windows, or Linux. Download now!
We are hiring!
If you're passionate about the topics we cover on our blog, please consider joining our team to help us ship the future of software development.
Instagram Engineering
Shopify Engineering