ollama-open-webui-local-install

name	ollama-open-webui-local-install
description	Install Ollama (local bare-metal) + Open WebUI (docker) on Ubuntu/Linux when official install script fails or sudo unavailable. Covers tmpfs /tmp space issues, tar structure quirks, Docker networking, and standalone docker-compose.
version	1.0.0
author	Hermes Agent
license	MIT
metadata	{"hermes":{"tags":["ollama","open-webui","docker","local-llm","installation"]}}

Ollama + Open WebUI — Local Install Guide

Context

Installing Ollama (bare-metal, local) + Open WebUI (docker) on Ubuntu when:

• Official curl -fsSL https://ollama.com/install.sh | sh fails (needs sudo password)
• /tmp is a tmpfs (RAM disk) with limited space — filling it breaks downloads
• docker compose plugin not available (use standalone docker-compose binary)
• Open WebUI in docker needs to connect to Ollama on host network

Architecture

Surface Pro 3 (Ubuntu):
  ├── Ollama (bare-metal)  →  localhost:11434
  └── Docker: Open WebUI   →  localhost:3000  →  host.docker.internal:11434

Step-by-Step

1. Download Ollama binary to ~/tmp (NOT /tmp)

mkdir -p ~/tmp
cd ~/tmp
curl -fSL "https://github.com/ollama/ollama/releases/download/v0.21.2/ollama-linux-amd64.tar.zst" -o ollama.tar.zst

⚠️ /tmp is a tmpfs (RAM disk). The ~2GB archive will fill it and cause "No space left on device" errors. Always use ~/tmp for large files.

2. Extract

cd ~/tmp
tar -I zstd -xf ollama.tar.zst
# Binary is at: ~/tmp/bin/ollama  (NOT ~/tmp/ollama)

3. Install to ~/bin (no sudo needed)

mkdir -p ~/bin
cp ~/tmp/bin/ollama ~/bin/ollama
chmod +x ~/bin/ollama
export PATH="$HOME/bin:$PATH"  # add to ~/.bashrc too

4. Start Ollama

nohup $HOME/bin/ollama serve > /tmp/ollama.log 2>&1 &
# Verify:
curl localhost:11434/api/tags

5. Pull a model (small, RAM-friendly)

$HOME/bin/ollama pull qwen2.5:1.5b   # ~1GB, good for 8GB RAM systems
# Or: $HOME/bin/ollama pull llama3.2:1b

6. Configure Open WebUI (docker)

Create ~/open-webui/.env:

OLLAMA_DOCKER_TAG=latest
WEBUI_DOCKER_TAG=main
OPEN_WEBUI_PORT=3000
WEBUI_SECRET_KEY=your-secret-key
OLLAMA_BASE_URL=http://host.docker.internal:11434
OLLAMA_API_BASE_URL=http://host.docker.internal:11434/api

Create ~/open-webui/docker-compose.yaml:

services:
  open-webui:
    image: ghcr.io/open-webui/open-webui:${WEBUI_DOCKER_TAG-main}
    container_name: open-webui
    volumes:
      - open-webui:/app/backend/data
    ports:
      - ${OPEN_WEBUI_PORT-3000}:8080
    environment:
      - 'OLLAMA_BASE_URL=http://host.docker.internal:11434'
      - 'OLLAMA_API_BASE_URL=http://host.docker.internal:11434/api'
      - 'WEBUI_SECRET_KEY=${WEBUI_SECRET_KEY}'
    extra_hosts:
      - host.docker.internal:host-gateway
    restart: unless-stopped

volumes:
  open-webui: {}

7. Start Open WebUI

Find docker-compose (standalone):

which docker-compose  # usually ~/.local/bin/docker-compose on systems without docker compose plugin
/home/marco/.local/bin/docker-compose up -d

Wait ~60s for initialization, then:

curl http://localhost:3000/api/version  # should return {"version":"0.x.x",...}

8. Auto-start Ollama (optional systemd user service)

mkdir -p ~/.config/systemd/user
cat > ~/.config/systemd/user/ollama.service << 'EOF'
[Unit]
Description=Ollama Local LLM Server

[Service]
ExecStart=/home/marco/bin/ollama serve
Restart=always
RestartSec=10

[Install]
WantedBy=default.target
EOF

systemctl --user daemon-reload
systemctl --user enable --now ollama

Troubleshooting

Ollama serve fails to bind port

pgrep -a ollama  # check if already running
pkill -f ollama  # kill existing instance

Docker can't reach Ollama

• Verify extra_hosts: host.docker.internal:host-gateway in compose
• From inside container: curl http://host.docker.internal:11434/api/tags

Open WebUI API returns "Not authenticated"

• Normal — UI requires auth but API endpoints are available
• Register at http://localhost:3000 for first account

docker-compose command not found

# Find standalone docker-compose
find ~ -name "docker-compose" -type f 2>/dev/null
# Or install: pip install docker-compose

Verification Checklist

• curl localhost:11434/api/tags → shows installed models
• docker ps → open-webui status "healthy"
• curl localhost:3000/api/version → returns version JSON
• Open http://localhost:3000 in browser → WebUI loads