Everything Eaon does.
The home page has the highlights. This is the long version: each part of the app, what it is for, and why it works the way it does. If it is on this page, it shipped.
Every way to run a model.
Hosted for you, on a key you already pay for, or on your own hardware. Same window either way.
Every model, one menu
The built-in connection serves the current line-up (Claude, GPT, Gemini, Llama, Mistral, DeepSeek, Grok, Qwen and the rest) and Eaon puts them all in a single picker. Switch in the middle of a conversation and the next reply just comes back from the model you chose.
Or skip the cloud
Point Eaon at Ollama, llama.cpp or MLX and the model runs on your laptop. It finds the runtimes you have already installed, and pulls new ones from Hugging Face or Ollama. You choose the quantisation and it fetches the right file.
Bring your own key
Already paying OpenAI, Anthropic or Google? Add your key and Eaon talks to them from your machine, with nobody in between.
Turn providers on and off
A coarse switch per connection, right in the model picker. Quiet down the built-in models and keep your own Anthropic key, or the other way round.
Once it is the window you leave open.
These are the parts that start to matter after the first week or two.
Ask once. Get two answers.
Put two models side by side and send the same prompt to both. You stop guessing which one suits a job and read the difference.
⌘K and go
One shortcut opens a command bar. Jump to any conversation, switch the model, flip the theme, or drop straight into a settings page, all with the arrow keys and Enter.
Projects
Group related chats into folders. A project's chats stay tucked inside it instead of piling into your recent list, so a week of digging into one thing doesn't bury everything else.
Custom instructions
Tell every model how you like your answers once, and it sticks. Leave it blank and nothing shapes the replies but your own message, which is how it ships.
Pinned chats
Keep the conversations you keep coming back to pinned at the top, sitting above the usual by-date list so they're one click away.
Search
Find any chat by what is inside it, project ones included. It shares a box with the command palette, so it is one shortcut away.
The reply itself got sharper.
How an answer lands on screen changes how much use you get out of it.
Code that arrives in colour
When a model writes code, you watch it appear syntax-highlighted, line by line, not as a grey block that only lights up once it's finished. And if you've scrolled up to re-read something earlier, Eaon leaves you there.
Memory that keeps up
Switch it on and Eaon picks up the specific things you tell it, over weeks, instead of holding one static profile. It keeps what you approve and nothing else. Off until you turn it on, and yours to wipe.
Reasoning you can fold away
Ask a reasoning model and its chain-of-thought folds into a collapsed "Thinking" dropdown under the answer. Open it when you are curious. Leave it shut and you get the reply, without a screenful of raw think tags in front of it.
One reply, however many steps
When a model makes a tool call and then keeps going, it renders as one continuous message instead of restarting the header and typing dots for every step. A quiet line between steps tells you what's happening right now.
Always clear who answered
Every reply carries a small header with the model's name, so a thread where you switched models halfway through still tells you who wrote which paragraph.
Make it do more than chat.
Reach your own tools and services without leaving the window.
MCP servers and a skill library
Paste a token and connect any custom MCP server, so a model can reach the tools and data behind it. Install a skill library straight from GitHub, then call any enabled one in the chat by typing /skill-name.
Make pictures, too
Generate images without leaving the app, through a hosted image model, your own cloud key, or a diffusion server on this Mac. No extra setup.
Your own local API server
Flip on a local, OpenAI-compatible server and point any external tool at it: a script, a coding CLI, another app. Bound to this Mac, and it wants an API key by default.
Tool calls without the nagging
Code execution and connected-plugin tool calls can run without a prompt each time. Device Control is the one exception. It can move your mouse and type for you, so it asks before every action whatever else you have set.
No surprises at the end of the month.
Statistics & cost
Every message is counted: tokens in and out, a running cost estimate per model, and where your usage is heading. Hosted models get a figure. Local ones cost nothing to run, and the panel says so instead of printing a zero you have to interpret.
Context-window meter
A small gauge shows how much of a model's context window the conversation is using, live. Where Eaon cannot read the real limit, the gauge stays empty instead of guessing at one.
It lives on your machine.
You can check every one of these yourself.
Keys in the Keychain
Eaon puts every API key in the macOS Keychain, the built-in one and your own, beside the rest of your passwords. No plain-text file, and no copy on our servers.
History is just files
Conversations are saved as files on your disk. Export any of them to Markdown or JSON, import them back later, or wipe the lot from Privacy settings.
Offline when you want it
With a local model loaded, nothing leaves the Mac, not the prompt, not the reply. Close the Wi-Fi and keep going; it won't notice.
It feels like a Mac app because it is one.
Signed, and built for both chips
A Universal build that runs on Apple silicon and Intel, signed and notarized, so it opens on a double-click. 2026.5.0 reworked streaming, so long replies stay smooth.
Light, dark, and yours
Monochrome by default, following your system theme, with a few accent options if you want a little colour. Both themes are designed together.
A typeface with a point of view
Space Grotesk on a fresh install, and every font on your Mac in the picker. One family throughout, so the whole thing reads as one piece.
Updates that find you
When there's a new version, a small card slides in. One click downloads it. You won't have to come back to this site to stay current.
A gentle start
First launch walks you through connecting a model and finding your way around, then gets out of the way and lets you get on with it.
Small, honest details
Keyboard navigation throughout, tidy empty states, and copy that tells you what is happening. You notice these when they are missing.
Get Eaon.
Free and open source. Your keys, your models, and your history stay on your machine.
Eaon for Windows isn't code-signed yet, so SmartScreen will show a blue "Windows protected your PC" box on first run: click More info, then Run anyway. That's expected, not a sign anything's wrong. Release notes ↗
Universal · macOS 14 Sonoma or later · 2026.5.0 · 0 downloads