Cursor alternatives — local AI for coding on a Mac

Cursor is a VS Code fork with serious AI baked in: in-editor chat, multi-file edits, predictive completion. The catch is it's cloud-backed. Maybe you care about privacy. Maybe it's the cost, or maybe you just work somewhere the cloud doesn't reach. If you want that same workflow running on a local model, these are your real options on a Mac.

What Cursor actually does

Strip away the branding and three things make Cursor feel different from plain VS Code:

1. A chat sidebar that's always there and knows your project.
2. Multi-file diff-based edits ("change this and update all the call sites").
3. Predictive completion that goes well past next-token autocomplete.

The first two come down to model quality, not the editor. The third comes down to latency. Cursor's predictive completion runs on fast cloud models tuned for sub-100ms first-token latency, and that's the property nobody has cracked locally yet.

The honest local options on a Mac

Setup	Editor	Local model	How Cursor-like
Continue.dev + Ollama	VS Code / JetBrains	llama.cpp via Ollama	Chat + multi-file edits; weak predictive completion
Continue.dev + LM Studio	VS Code / JetBrains	llama.cpp / MLX	Same as above
Outlier (separate app)	Outlier app, not in editor	Core 27B / Code 27B / Plus 397B	Full agent workflow outside the editor
Zed + local LLM	Zed editor	Configurable LLM API endpoint	Lighter integration, growing fast

What each tradeoff looks like

Continue.dev (in-editor)

If you want Cursor's feel without leaving the editor, this is it. Sidebar chat, slash commands, multi-file edits, all wired straight into VS Code or JetBrains. Point it at Ollama or LM Studio for the local backend and you're set. Predictive autocomplete works. It's just noticeably slower than Cursor's cloud version, because the model lives on your machine.

Outlier (separate app)

Outlier isn't an editor plugin. It's a Mac-native AI app, and the agent loop is the whole point. It reads your project, edits files, runs commands, and pauses for approval when it should. You hop between your editor and Outlier instead of getting AI inline. So it's less "AI in your face" and more "an agent quietly doing structured work." Core 27B's HumanEval pass@1 sits at 0.8659. That's high enough for the local agent to hold its own on real refactors.

Zed

Zed is the new kid. Lightweight, and you can point it at any LLM API endpoint, a local Ollama or LM Studio included. The integration is thinner than Cursor or Continue.dev. But the editor is genuinely fast, and the direction it's heading is worth watching.

The predictive-completion gap

This is the part where local really shows its seams. Cursor's "tab completion" fires off multi-line, context-aware suggestions in under 100ms. Matching that on your own machine takes two things:

• A small, fast model (Outlier Nano 4B is in the right size class) that can spit out a multi-line completion in under 100ms.
• An editor integration that shows it without flicker.

Continue.dev pulls this off with small local models, but the experience is rougher than Cursor's cloud version. Predictive completion is the one spot where Cursor's cloud edge is real and you feel it every day.

Picking your setup

• You want as close to Cursor as possible, in-editor → Continue.dev + a local backend (Ollama, LM Studio, or Outlier's HTTP endpoint).
• You want a real local coding agent and don't mind leaving the editor → Outlier app.
• You want a fast modern editor with optional local AI → Zed.
• You need predictive completion as good as Cursor's → stay on Cursor for now, or live with the gap. No shame in either.

Frequently asked questions