home-lab · self-hosted · GPL-3.0

One master.
Every AI host
in your home network.

octollama is a web panel that manages Ollama, the LiteLLM gateway and Open WebUI across many machines at once — from a browser, phone included. Add hosts, pull models, wake sleeping computers, expose everything behind a single OpenAI-compatible endpoint.

View on GitHub Get started
live demo

Watch how it works

A 70-second loop: the master, attaching slave hosts, one gateway for every model, Wake-on-LAN and automatic failover.

features

What the panel does

🦙

Service control

Start, stop and autostart Ollama on every host; tune each performance variable — context size, VRAM, Vulkan/iGPU, KV cache.

📦

Model management

See what is installed and what sits in memory, pull new models with a progress bar, delete what you no longer need.

🔀

LiteLLM gateway

One OpenAI-compatible endpoint over the models of every host. You decide exactly which models get exposed.

💬

Open WebUI

A chat wired to the gateway — it sees the same, deliberately selected models from every host at once.

🧩

Continue.dev config

VS Code config generated from the currently exposed models, ready to paste. The panel never overwrites your file.

🖧

Multi-host

Add a remote machine and the panel generates a ready-to-run installer for it. Run once — the host joins the network.

⚡

Power management

Wake-on-LAN for sleeping hosts plus remote power off, restart and suspend — straight from the panel.

🔐

No root in the web

The panel never touches systemctl. A local root daemon with zero network ports applies changes through a state file.

under the hood

Security by design

Changing systemd services needs root — so the web panel doesn't get it. The panel writes the desired state to state.json; a separate root-owned daemon watches the file, applies the change and reports back in status.json. The daemon's only contact with the outside world is a file on disk. Remote hosts mount their state directory over NFS, exported per host and restricted to its IP.

install

Up and running in minutes

On the machine that will become the master (Debian/Ubuntu, Python 3.11+):

git clone https://github.com/cyryllo/OctoLlama.git
cd OctoLlama
./install.sh

The installer sets up Ollama, LiteLLM, Open WebUI, the daemon and the panel — then the panel lives at http://<master>:5000. Free and open source, GPL-3.0.