Outlier vs Enchanted: bundled Mac app vs open-source Ollama client

Disclosure up front: I'm Matt Kerr, I build Outlier, and I'm the interested party here. I've tried to write this so that someone who ends up choosing Enchanted still finds it useful. Enchanted is genuinely good software, it's free, and it does several things Outlier can't do at all.

Privacy isn't the difference here

Most comparisons on this site pit a local app against a cloud service, and privacy carries the whole argument. Not this one. Point Enchanted at an Ollama server on your own Mac and your prompts never leave the machine — same as Outlier. Its App Store privacy label says the developer collects no data, its dependency list contains no analytics SDKs, and the macOS build is sandboxed with four entitlements (network client, microphone, read-only access to files you pick). That's structural, not promissory, because there's no vendor backend to send anything to.

One caveat that cuts both ways: Enchanted talks to whatever endpoint you type in. If that endpoint is a hosted Ollama somewhere else, your prompts go there. And the README's own recipe for reaching a home server from your phone is an ngrok tunnel — ngrok http 11434 — which puts your model behind a publicly reachable third-party URL. That's a real hole in an otherwise airtight story, and it's the project's own documented method.

Client versus bundle: the actual split

Enchanted performs no inference. The README is blunt about it: you need to run your own Ollama server, version 0.1.14 or later. The app is 7.2 MB because there's nothing in it but the interface. You install Ollama separately, pull models with Ollama's command line, keep both updated, and troubleshoot the server when something goes wrong. Enchanted is the pretty ChatGPT-shaped window on top.

Outlier ships the other way. The runtime and the model catalogue are the product: you download the app, pick a tier, and it fetches the weights itself. No server process, no terminal, no version pinning. Nano (4B, 2.37 GB, 6 GB RAM) and Lite (9B, 5.04 GB, 12 GB RAM) are free with no account. Pro adds Quick, Core, Code, Vision and Plus — Core 27B runs about 20.7 tok/s on an M1 Ultra, and Plus 397B streams its experts from SSD to fit in roughly 11 GB of peak RSS on a 64 GB machine.

Which is better depends entirely on whether "install and manage Ollama" reads to you as a five-minute chore or a wall. If you already run Ollama, Enchanted costs you nothing and adds a good interface to something you have. If you don't, Enchanted's setup is Ollama's setup plus an app.

Who each one is for

Enchanted's own positioning answers this. The App Store listing was renamed "Enchanted Developers Only" in version 1.9.0, with the entire release note reading "App name changed to reflect that it's for developers only." It sits in the Developer Tools category and its description talks about LLM researchers chatting with self-hosted models. That's an unusually honest piece of self-labelling, and it tells you who should walk in.

Outlier is built for the person who wants a chat app and has no interest in the machinery underneath. No account, no caps, no server, works with Wi-Fi off. Different audiences, and I don't think either framing is wrong.

Side by side

Where Enchanted genuinely wins

Several of these aren't close.

• It's free and it's open. Apache-2.0, roughly 6,000 GitHub stars, the entire Swift source public. You can read exactly what it sends where, fork it, or build it yourself. Outlier's app is closed source — you have to take my word for the privacy claims, or watch the network traffic. Only our weights are public.
• Four Apple platforms, one download. macOS, iPhone, iPad and Vision Pro. Outlier is Mac-only and Apple Silicon only, with no mobile app of any kind. If you want the same conversations on your phone, Outlier simply cannot do it, and that's a real disadvantage.
• Splitting the UI from the runtime is a genuinely better architecture for some setups. Run a big model on a desktop with lots of RAM and drive it from a thin client. An app that does inference on-device is hard-capped by the RAM of the machine it's installed on — Outlier's Plus tier needs 64 GB, full stop.
• Model-agnostic. Anything in the Ollama library works, including models released long after the app stopped being updated. You can switch models mid-conversation or resubmit an edited message to a different model, which is a fast A/B workflow. Outlier gives you seven tiers I picked.
• Nice Mac-native touches. A Ctrl+Cmd+K Spotlight-style panel to prompt from anywhere, prompt templates ("Completions") that act on selected text in other apps, read-aloud with system voices, voice prompts, image attachments, proper markdown and code rendering.
• 7.2 MB. Install and uninstall cost is basically nothing.

Where Outlier has an edge

It's self-contained: one download, no Ollama, no terminal, no endpoint URL to configure, and the models come with it. It's actively maintained — Enchanted's last functional code commits date to March 2025, the App Store build hasn't moved since August 2025, 113 issues sit open, and and the README's first line now points to Jaz as "the new iteration of this project" — a different product entirely. Enchanted still works; it just isn't going anywhere.

On output quality I can only speak to Outlier's side, since Enchanted's quality is whatever model you pulled. Our Core 27B matched Claude Opus on 98.9% of rubric checks across a 54-prompt head-to-head, and our local 27B measured about 45% on a blind slice of SWE-bench Verified. Also worth saying plainly: Outlier has no web search, no image generation, no voice mode, no cloud sync.

Who should pick which

Pick Enchanted if you already run Ollama, you want your models on an iPhone or a Vision Pro, you want to point one app at a server on another machine, you need open source you can audit or fork, or you want to swap freely among many models. It costs nothing to try.

Pick Outlier if you're on an Apple Silicon Mac, you don't want to install or maintain a server, you want the models handled for you, and you'd rather pay for something under active development than run a free tool whose author has moved on.

Skip both if you're on Windows or Linux. Neither app runs there.