Quo 1.7.1

What's new

• A newer engine. Quo now runs llama.cpp's first numbered stable release, v0.9.0. Replies came faster in our tests on the graphics card, Gemma 4 E2B reads long messages faster, and Quo can run 8 more kinds of model from Hugging Face, among them Mistral Large 4, LFM3 and DeepSeek 4.1.
• A picture to make. A new chat always suggests one picture among its ideas. Click it to make the picture: in an offline chat it's made on this computer once an image model is downloaded.
• Images made on your computer use the newest stable-diffusion.cpp.

Fixed

• Stable Diffusion 2 files: adding one as your own image model now says plainly that Quo can't run it, instead of treating it as Stable Diffusion 1.5.
• Security: the app is built on Electron 44.10, with Chromium's latest fixes.

Notes

• Your chats and downloaded models are kept.
• On a Mac, Quo tells you when a new version is out (it can't update itself yet). Windows and Linux (AppImage) update themselves.
• The builds aren't code-signed yet. macOS: right-click → Open the first time. Windows: More info → Run anyway. Or install in one line: curl -fsSL https://quo.computer/install.sh | sh (macOS and Linux) or irm https://quo.computer/install.ps1 | iex (Windows).
• Built by GitHub Actions from this repository. The chat engine is the official llama.cpp release v0.9.0 (b12110) and the image engine stable-diffusion.cpp master-1052-f5df5f8, each checked against its SHA-256.

Quo 1.7.0

What's new

• Images made on your computer. Download an image model in Models & storage (Z-Image Turbo, or FLUX.2 klein, which can also change a picture) and offline chats make pictures on this computer: nothing is sent, not even the description. Ask for a change right after a picture ("remove the apple") and Quo makes it. You can also add an image model file of your own: SDXL, Stable Diffusion 1.5, or a Z-Image or FLUX.2 klein fine-tune.
• Images. Every picture you've made, across chats, in the sidebar. Click one to see it full size, zoom in and move around.
• Tool servers (MCP). Give models more tools, such as your calendar, GitHub or a database. Add servers to Quo's tool servers file (the same form other apps use) in Models & storage → Code. Turning one on shows exactly what it runs, and Quo asks before each of its tools is used.
• Documents in a folder. Quo reads PDF, Word, PowerPoint, Excel and OpenDocument files in a folder you open, so you can ask about a folder of documents. They are read in a sealed-off part of the app.
• A local server for other apps. Let editors, scripts and tools on this computer use Quo's offline model through the OpenAI-style API. Off until you turn it on in Models & storage.
• Fewer questions in Code mode, if you choose. For one session, let Quo ask only about risky steps, or about nothing but pushes.

Improved

• Code mode finds things out with its tools instead of saying it can't, and with the free model it takes up to 20 steps a reply instead of 8.
• Pictures in offline chats are made on this computer once an image model is downloaded; otherwise Quo asks before making one in the cloud.

Fixed

• The next message after a picture loads the offline model again and gets an answer, also after the faster NVIDIA engine is installed or an unused model was let go.
• Quo's approval window shows again. Released builds had shown the system's plain dialog for every approval instead.
• The Plan button no longer runs over the note beside it.
• Linux one-line install: the menu entry matches the app, and quo:// links open in it.

Notes

• Your chats and downloaded models are kept.
• On a Mac, Quo tells you when a new version is out (it can't update itself yet). Windows and Linux (AppImage) update themselves.
• The builds aren't code-signed yet. macOS: right-click → Open the first time. Windows: More info → Run anyway. Or install in one line: curl -fsSL https://quo.computer/install.sh | sh (macOS and Linux) or irm https://quo.computer/install.ps1 | iex (Windows).
• Built by GitHub Actions from this repository. The chat engine is the official llama.cpp release b11719 and the image engine stable-diffusion.cpp master-1044-f8ea2e1, each checked against its SHA-256.

Quo 1.6.0

What's new

• A new look. A calmer design with solid surfaces and Atkinson Hyperlegible type, made to be easy to read. Choose Light, Dark or Black; Quo follows your system's high-contrast setting and e-ink screens. Every reply shows where it ran: on your device or in the cloud.
• Edit a message, or go back. Change one of your earlier messages and the chat goes on from there, or go back to an earlier reply, with a few seconds to undo. Quo Sync keeps the change on your other devices.
• Make images. Ask a cloud model for a picture, or type /image with any model. A few are free every day, then they use credits, and each image shows what it cost. Images stay on your computer.
• Files and photos for every model. Documents and photos now go with your message to cloud models and to models you use with your own key. With an offline model they still never leave your computer.
• Quo Pro, credit packs and Quo Sync are on sale under Plan & credits. New accounts get 25 free credits to try.
• Invite friends. A friend who signs up with your link gets 25 extra credits, and you get 50 when they first buy.
• Open a model from Hugging Face. quo://models/huggingface/<owner>/<name> links open that model in Models & storage, ready to download. A link never downloads anything by itself.

Notes

• Quo now uses its own addresses: app.quo.computer and quo.computer. The old address keeps working for a while.
• Your chats and downloaded models are kept.
• On a Mac, Quo can't update itself yet: from this version it tells you when a new one is out. Windows and Linux (AppImage) update themselves.
• The builds aren't code-signed yet. macOS: right-click → Open the first time. Windows: More info → Run anyway. Or install in one line: curl -fsSL https://quo.computer/install.sh | sh (macOS and Linux) or irm https://quo.computer/install.ps1 | iex (Windows).
• Built by GitHub Actions from this repository; the engine is the official llama.cpp release b11719, checked against its SHA-256.

Quo 1.5

What's new

• Any model from Hugging Face. Search Hugging Face in Models & storage and run almost any GGUF model: about 160 kinds, including Llama, Qwen, Gemma, Mistral, DeepSeek, GLM, Phi and Granite. Quo tells you before downloading if a model won't work, and suggests a size that fits your computer.
• Faster writing. Qwen and Gemma models draft several words at a time and check each one, so replies are the same, only quicker. Rewriting a file with a small change went from about 100 to 570 words a second in our test.
• Faster on NVIDIA graphics cards. Models & storage offers llama.cpp's CUDA engine: long messages are read about 40% faster and replies written about 10% faster. It's downloaded only if you ask (550–770 MB), from llama.cpp's own release, and checked before use. If it ever fails to start, Quo goes back to the engine it came with.
• Much faster follow-ups. The model reuses what it already read of the chat instead of reading it all again: in one test, a follow-up took 2 seconds instead of 114.
• Longer memory on smaller computers. When memory is short, the model's working memory is stored in 8 bits, which fits about twice as much conversation.
• New models. Qwen3.8 27B replaces Qwen3.5 27B; Gemma 4 12B and 26B A4B join Gemma 4 E2B, each with a small drafter file for faster writing.
• Linux on ARM. New Quo-linux-arm64 AppImage and .deb.
• Dark mode and themes, code colours and maths. True black or Midnight, an Indigo or Mono accent, code coloured by language, and formulas drawn properly.

Fixes

• In Safari and older browsers, Quo Web no longer loads part of its engine from a public CDN.

Notes

• Your chats and downloaded models are kept. Gemma 4 E2B moves to a smaller, better file; the one you have still works.
• If you had Qwen3.5 27B, Models & storage offers to delete its files (about 18 GB).
• The builds aren't code-signed yet. macOS: right-click → Open the first time. Windows: More info → Run anyway.
• Built by GitHub Actions from this repository; the engine is the official llama.cpp release b11719, checked against its SHA-256.

Quo for desktop 1.4

What's new

• Push and pull requests. Code mode can push the branch you're on and give you a link to open a pull request on GitHub. You approve every push and see the commits first; it never force-pushes. Turn it off in Models & storage → Code.
• Code mode starts itself. Ask for something in a project and the model asks for the folder (you approve it). It can also run a dev server in the background, keep a checklist, and summarise long sessions.
• Searches and sources. The model searches the web when a question needs it, with no switch to press, and the pages it used appear under the reply. Quo's free model gets 3 searches a day.
• Faster offline Qwen3.5. New downloads predict several words at a time, about twice as fast on code edits. Models that don't fit your memory start with a smaller window instead of failing.
• Thinking levels. Off, Quick, Auto or Deep, showing only the levels each model has.
• Approvals in Quo's own window, a file panel with clickable file links, and a memory meter for long chats.

Fixes

• macOS: Code mode refuses system folders reached through links.
• Linux: the taskbar shows Quo's icon.

Notes

• Your chats and downloaded models are kept. Qwen3.5 files you already have still work; download them again in Models & storage for the faster version.
• Qwen3.5 0.8B and 2B are no longer offered in the desktop app (too small for good answers); Models & storage offers to delete their files.
• The builds aren't code-signed yet. macOS: right-click → Open the first time. Windows: More info → Run anyway.
• Built by GitHub Actions from this repository; the engine is the official llama.cpp release b11719, checked against its SHA-256.

Quo for desktop 1.3

What's new

• Code mode. A coding agent for your project folders: turn on Code in the chat bar. It reads and searches the code, makes changes and runs tests, and you approve every change and command. Plan mode makes it read-only, each reply that changed files has an Undo button, and it follows your project's AGENTS.md. It works with Quo's cloud models, your own API key, or an offline model on your computer (Qwen3.5 9B and up recommended).
• Use your own API key. OpenRouter, Anthropic, OpenAI, Google, xAI, DeepSeek, Mistral, Groq, or a server of your own such as Ollama, in Models & storage → Your keys. No Quo account, and messages go straight to the provider. Your key is encrypted by your system's secure storage and the app's pages can't read it back.

Notes

• Your chats and downloaded models are kept.
• The builds aren't code-signed yet. macOS: right-click → Open the first time. Windows: More info → Run anyway.
• Built by GitHub Actions from this repository; the engine is the official llama.cpp release b11719, checked against its SHA-256.

Quo for desktop 1.2

What's new

• Memories. Things you want Quo to know about you, kept under Memories in the sidebar: "I'm vegetarian", "my name is Sam". Offline models read them before every reply, so you don't have to repeat yourself. They are never sent to a cloud model.
• Sign out of all devices. For a lost phone or a shared computer: one button ends every other sign-in on your account. It appears with accounts (see below).
• Quo Sync now covers the Android app too. Chats and memories, end-to-end encrypted with a recovery key only you hold. Built in, but off until accounts are switched on.

Notes

• Your chats and the models you downloaded are kept. The chat database is upgraded in place to hold memories (tested against a real 1.1 database).
• The builds aren't code-signed yet. macOS: right-click → Open the first time. Windows: More info → Run anyway.
• Windows, Linux and macOS builds are produced by GitHub Actions from this repository; the engine is the official llama.cpp release b11719, checked against its SHA-256.

Quo for desktop 1.1

What's new

• A native AI engine. Models now run in llama.cpp on your graphics card (Metal on Apple silicon, Vulkan on Windows and Linux) or your processor, instead of inside the browser. On a Core i7-9700K with an RTX 3070 Ti, a small model writes about 250 tokens a second instead of about 12, and a long document is read in under a second instead of over a minute. There is no 2 GB model limit any more.
• More models. Qwen3.5 up to 27B, Gemma 4, Llama 3.2 and Phi-4 mini, and models that read photos. Quo recommends the biggest one that fits your computer.
• Models you already have. Use files from LM Studio, Ollama or any GGUF where they are, with no second download.
• Photos and documents. Attach PDFs, text files and (with models that read images) photos. They're read on your computer and only go to offline models.
• Voice. Talk instead of typing, and have replies read aloud. Speech is turned into text on your computer, after a one-time download of about 80 MB.
• If a computer isn't supported by the native engine, Quo falls back to the in-browser engine.

Quo Sync (end-to-end encrypted chat sync) is built in but off until accounts are switched on.

Notes

• The builds aren't code-signed yet. macOS: right-click → Open the first time. Windows: More info → Run anyway.
• Your chats and the models you downloaded in 1.0 are kept.
• Windows, Linux and macOS builds are produced by GitHub Actions from this repository; the engine is the official llama.cpp release b11719, checked against its SHA-256.

Quo for desktop 1.0

The first desktop release of Quo AI: private AI that runs on your computer and works offline.

Download

• macOS (Apple silicon and Intel): Quo-mac-universal.dmg
• Windows 10 and 11: Quo-windows-x64-setup.exe
• Linux: Quo-linux-x86_64.AppImage or Quo-linux-amd64.deb

What's in it

• Chat with open models (Qwen3.5, Llama 3.2) that run on your graphics card
• Download a model once; after that Quo works with no internet, and nothing you type leaves your computer
• Thinking mode, saved chats and search, markdown answers
• Quo suggests a model that fits your computer
• Optional free cloud model, clearly marked and only used when you pick it
• Updates itself on Windows and Linux

First launch
These builds aren't code-signed yet:

• macOS: if it says it can't check the app, open System Settings › Privacy & Security and choose Open Anyway. Updates on macOS are manual until the app is signed.
• Windows: if you see "Windows protected your PC", choose More info › Run anyway.
• Ubuntu 22.04+: the AppImage needs libfuse2, or use the .deb.

For developers
Quo for desktop is Electron serving the Quo Web build (apps/web) from a cross-origin isolated app:// origin, so WebGPU and multi-threaded WebAssembly are always available. Source and build instructions: apps/desktop.