A small, dark-themed desktop app for Linux that gives Ollama a proper window: see your models, chat with them, tweak their settings and install or remove models — without the terminal.
Ollama's own desktop app doesn't support Linux, so this fills that gap. It is not a fork of Ollama
and it doesn't bundle it: it simply talks to the Ollama server that's already running on your computer
(http://127.0.0.1:11434) through its normal HTTP API.
Unofficial personal / learning project. Not affiliated with or endorsed by the Ollama project.
- Model list — every installed model with its size, parameter count and when it was last changed. Models currently loaded into memory get a "loaded" badge. If Ollama isn't running you get a clear message, the commands to start it, and a Retry button (it also reconnects by itself).
- Chat with streaming replies — words appear as the model produces them. Pick the model from a drop-down (or the sidebar), Stop a reply half-way, or Clear chat to start over.
- Per-model settings that are remembered — temperature, top P, context length and a system prompt,
saved separately for each model and restored next time you open the app. They're sent with each chat
request as Ollama
options. Stored in~/.config/ollama-desktop/settings.json. - Install models by name (e.g.
llama3.2,qwen2.5:7b) — asks for confirmation first, then shows a live progress bar with downloaded / total size. Cancel any time; Ollama resumes where it left off. - Uninstall models — with a confirmation dialog that tells you how much disk space you'll get back.
Privacy & safety: the app only ever connects to Ollama on this machine (127.0.0.1). It has no accounts,
no telemetry and no API keys, and it never changes how Ollama itself is exposed to your network.
- Linux (developed for Ubuntu; other distros should work).
- Ollama installed and running. Install it from https://ollama.com/download/linux:
The installer sets Ollama up as a service. If it isn't running, start it with
curl -fsSL https://ollama.com/install.sh | shsudo systemctl start ollama(or runollama servein a terminal). - Node.js 20.19 or newer (Node 22 LTS recommended) and npm — only needed to run from source or build.
Check with
node -v.
git clone https://github.com/samkelly880/ollama-desktop.git
cd ollama-desktop
npm install
npm run devnpm run dev opens the app with hot reload: edits to the UI (src/renderer) show up immediately;
changes to src/main or src/preload need a restart (Ctrl+C, then npm run dev again).
Newer Ubuntu versions restrict the kernel feature Electron's sandbox uses, so running Electron straight
from node_modules can fail with "The SUID sandbox helper binary was found, but is not configured
correctly". Fix it once per npm install with:
sudo chown root:root node_modules/electron/dist/chrome-sandbox
sudo chmod 4755 node_modules/electron/dist/chrome-sandboxor, for development only, start without Chromium's sandbox: npm run dev:no-sandbox.
The installed .deb doesn't need either: it sets up an AppArmor profile for you.
npm run buildThis type-checks, builds the app, and packages it with electron-builder into release/:
release/ollama-desktop-<version>-x86_64.AppImage— a single file you can run anywhere:On Ubuntu 22.04+ AppImages need FUSE 2:chmod +x release/ollama-desktop-*.AppImage ./release/ollama-desktop-*.AppImage
sudo apt install libfuse2(on 24.04:libfuse2t64). If it exits with a sandbox error on Ubuntu 24.04, start it with--no-sandboxor use the.debinstead.release/ollama-desktop_<version>_amd64.deb— installs into your app menu as "Ollama Desktop":Remove it again withsudo apt install ./release/ollama-desktop_*_amd64.debsudo apt remove ollama-desktop.
npm run build:dir makes just an unpacked folder (release/linux-unpacked/), which is quicker for testing.
| Command | What it does |
|---|---|
npm run dev |
Start the app in development mode |
npm start |
Build and run the production version (without packaging) |
npm test |
Run the unit tests (Vitest) |
npm run typecheck |
TypeScript type-check |
npm run lint |
ESLint |
npm run format |
Format everything with Prettier |
npm run check |
Type-check + lint + format check + tests (run before committing) |
src/
main/ Electron main process: window, IPC handlers, settings file
preload/ Exposes a small, typed `window.api` to the UI (context isolation, sandboxed)
renderer/ React UI
shared/ Code used by both sides: Ollama API client, settings rules, formatting
tests/ Vitest tests (the Ollama API is mocked; no real server needed)
All Ollama requests are made by the main process (src/shared/ollama.ts), which refuses any address
that isn't this machine. The UI asks for things over IPC; streaming responses (chat tokens, download
progress) are forwarded line by line as Ollama sends them. Endpoints used: /api/version, /api/tags,
/api/ps, /api/chat (streaming), /api/pull (streaming) and /api/delete.
Built with Electron, TypeScript, Vite (via electron-vite), React, electron-builder and Vitest.
- Render Markdown and code blocks in replies (with a copy button)
- Save chat history and keep several conversations in a sidebar
- Show tokens-per-second and other stats after each reply
- Search / browse the Ollama library from inside the app
- Show model details (
/api/show): licence, template, parameters, capabilities - "Unload from memory" button for loaded models (
keep_alive: 0) - Light theme and a theme toggle
- Optional custom Ollama port (still localhost-only)
- Keyboard shortcuts (new chat, focus input, switch model)
- Export a conversation as Markdown
- ARM64 builds and a Flatpak