Lume — Local AI Desktop Client

A fast, private AI chat client for Linux. Talks to Ollama running on your machine. No cloud. No telemetry. No Electron.

---

What it does

• Chat with any Ollama model installed on your system
• Full streaming output — tokens appear as they generate
• Multiple independent chat sessions with persistent SQLite history
• Markdown + syntax highlighting in responses
• Dark/light mode, collapsible sidebar, keyboard shortcuts
• Intelligence Governor — warns before loading a model that will OOM your system
• Offline detection — banner appears within 5 seconds if Ollama goes down

What it doesn't do

• No cloud. No accounts. No API keys.
• No GPU required (Ollama handles that separately)
• No Electron. WebKit via Tauri. ~5 MB binary, <20 MB idle RAM.

---

Requirements

	Minimum
OS	Linux (x86_64) — tested on Fedora 40, Ubuntu 24.04
RAM	8 GB system RAM (Lume uses <20 MB — the model uses the rest)
Disk	10 MB for Lume + space for Ollama model weights
Ollama	Must be installed and running before launching Lume

RAM by model size (rough guide): | Model | RAM needed | |---|---| | 1B–3B params | 2–4 GB | | 7B params | 6–8 GB | | 13B params | 10–14 GB | | 70B params | 40+ GB |

---

Install Ollama first

curl -fsSL https://ollama.com/install.sh | sh
ollama serve          # start the server
ollama pull llama3.2  # download a model

---

Install Lume

AppImage (any distro)

chmod +x lume_0.5.5-beta_amd64.AppImage
./lume_0.5.5-beta_amd64.AppImage

Debian/Ubuntu (.deb)

sudo dpkg -i lume_0.5.5-beta_amd64.deb

Fedora/RHEL (.rpm)

sudo dnf install ./lume-0.5.5-beta-1.x86_64.rpm

Arch Linux (AUR — community)

# AUR package coming soon

---

Build from source

Dependencies:

# Fedora
sudo dnf install webkit2gtk4.1-devel openssl-devel curl wget file \
  libappindicator-gtk3-devel librsvg2-devel pango-devel nodejs rust cargo

# Ubuntu/Debian
sudo apt install libwebkit2gtk-4.1-dev libssl-dev curl wget file \
  libappindicator3-dev librsvg2-dev libpango1.0-dev nodejs cargo

Build:

git clone https://github.com/sandeep4513m/Lume.git
cd Lume
npm install
npm run tauri dev             # dev mode with hot-reload
NO_STRIP=true npm run tauri build   # production build (.deb .rpm .AppImage)

---

Troubleshooting

Banner says "Ollama is offline"

ollama serve
# or if using systemd:
systemctl start ollama
systemctl enable ollama  # auto-start on boot

No models in the dropdown

ollama list              # check what's installed
ollama pull llama3.2     # pull a model

App won't start / blank screen

• Lume requires Ollama to be reachable at http://localhost:11434
• Custom Ollama URL can be set in Settings → Models

High RAM usage / system lag

• Open the System Health panel (⋯ menu → System Health)
• The Intelligence Governor will warn you before loading a model that exceeds available RAM
• Unload the current model before switching to a large one

Build fails on Ubuntu

# Make sure webkit2gtk 4.1 is installed, not 4.0
sudo apt install libwebkit2gtk-4.1-dev

---

Tech stack

Layer	Technology
Framework	Tauri v2 (Rust)
Frontend	Svelte 5
Styling	Tailwind CSS v3
Database	SQLite via rusqlite
Inference	Ollama REST API

---

License

MIT. See LICENSE.