Two native apps that put AI where you already work. Run them on your own keys, or on a model that stays on your hardware. Both are free and open source.
Universal for Apple silicon and Intel. Also on Windows and Linux.
Pick a model from one menu. Here it is a coding model running on this Mac through Ollama, answering without a network.
Models
Hosted on your own key, open-weight, or running on your own hardware, in a single list. Switch mid-conversation and the thread carries over.
OpenAI
Mistral
Qwen
Moonshot
Z.ai
xAIOr on your own machine, through a runtime you already have installed
Point Eaon at a running Ollama and every model it has pulled joins the picker.
Any GGUF you already have on disk, served straight from your own machine.
Apple silicon inference, so a Mac runs the model on its own GPU.
Eaon
Every major model in one window, on your key or on your own hardware.
Hosted models, providers on your own key and local ones share one picker. Switch mid-thread and the next reply comes from the model you chose.
Download a model from the Models page and it runs on your machine, through Ollama, llama.cpp or MLX. No network, no telemetry, no account.
Ask about something that happened this week. It searches, reads the pages, and every step shows up in the thread.
The search above the sidebar finds a chat by its title or by anything said in it, and opens it where you left off.
Switch to Work and it reads, edits and runs things in your project. Before any command runs, it stops and asks.
Pick a palette, or set the background, accent and contrast yourself. Every surface re-tones from those few values.
Eaon ADE
Claude Code, Codex, Gemini CLI, Aider, OpenCode, Copilot, Mistral Vibe, Kimi Code and Grok, each in a real terminal, side by side.
Type once. Send it to the pane you are in, to every pane, or to one you name.
Hold Right ⌘ and speak. A Whisper model on your own CPU writes it onto the prompt. No audio leaves the machine.
One colour means an agent is working. The other means you are the bottleneck, and it is used for nothing else.
Pick how many panes and which agent, write one opening prompt, and every pane starts on it at once. Compare what comes back.
Park the work you have not started, then drag each card along as it goes from queued, to an agent, to done.
It finds the dev servers that are actually up and shows them as one-click chips, next to the agent building them.
How it works
No installer wizard, no account, and no trial clock. The longest part is the download itself.
Signed and notarized, so it opens on a double-click the way any installed app does.
Use the built-in connection, paste your own provider key, or point it at a model running on your machine.
Switch models mid-conversation and the thread carries over. History stays on your disk as files.
The same models and the same keys, in a terminal agent that can read and write files in your project.
npm install -g eaon-cli
Questions
Yes. The app has no paid tier and no account to create. You pay whichever model provider you pick, or you pay nobody and run a model on your own hardware.
Only for hosted models. Local models need no key and no connection. You can use both in the same conversation.
On your machine. Cloud sync stays off until you turn it on, and it encrypts everything before upload, so the server holds data it cannot read.
No. The app is signed and notarized, so it opens the way any other installed app does.
Both apps and the CLI are on one page. None of them ask you to make an account, and none of them cost anything.