Skip to content

OLLAMA_INSTALL.md

Latest commit

 

History

History
184 lines (121 loc) · 3.31 KB

OLLAMA_INSTALL.md

File metadata and controls

184 lines (121 loc) · 3.31 KB

Ollama Installation Guide (macOS, Windows, Linux)

Overview

This guide shows how to install and run Ollama for Audify local script generation.

Recommended starter model:

  • qwen3:1.7b

Prerequisites

  • At least 8 GB RAM (16 GB recommended for smoother local inference).
  • ~4 GB free disk space for runtime and model files.
  • Docker Desktop installed if you run Audify via Docker Compose.

macOS Installation

1. Install Ollama

Option A (Installer):

  1. Download from https://ollama.com/download
  2. Open installer and complete setup.

Option B (Homebrew):

brew install ollama

2. Start Ollama service

ollama serve

Keep this terminal running, or run Ollama as a background app from the installer.

3. Pull a model

ollama pull qwen3:1.7b

4. Verify

ollama list
curl http://localhost:11434/api/tags

Windows Installation

1. Install Ollama

  1. Download the Windows installer from https://ollama.com/download
  2. Run installer and complete setup.

2. Start Ollama

Ollama usually starts as a background service after installation.
If needed, start from Start Menu: Ollama.

3. Pull a model (PowerShell)

ollama pull qwen3:1.7b

4. Verify (PowerShell)

ollama list
curl http://localhost:11434/api/tags

If curl aliases to Invoke-WebRequest, this also works:

Invoke-RestMethod http://localhost:11434/api/tags

Linux Installation

1. Install Ollama

curl -fsSL https://ollama.com/install.sh | sh

2. Start service

ollama serve

For systemd environments, you can run as a service if configured by your setup:

sudo systemctl enable ollama
sudo systemctl start ollama
sudo systemctl status ollama

3. Pull a model

ollama pull qwen3:1.7b

4. Verify

ollama list
curl http://localhost:11434/api/tags

Configure Audify to Use Ollama

Update api/llm-service/.env:

OLLAMA_BASE_URL=http://host.docker.internal:11434/v1
OLLAMA_MODEL=qwen3:1.7b

Notes:

  • host.docker.internal works for Docker Desktop on macOS/Windows.
  • On Linux, you may need to expose host networking differently or set OLLAMA_BASE_URL to your host IP.

Quick Functional Test

Run a local generation test:

curl http://localhost:11434/api/generate -d '{
  "model": "qwen3:1.7b",
  "prompt": "Write one sentence about document to podcast conversion.",
  "stream": false
}'

If you receive a JSON response with generated text, Ollama is ready.

Troubleshooting

Ollama command not found

  • Reopen terminal after install.
  • Confirm binary path is in PATH.

Port 11434 not reachable

  • Check service is running: ollama serve.
  • Check firewall settings.

Model pulls are slow

  • First pull downloads full model weights.
  • Verify network connectivity and disk space.

Audify cannot connect from Docker

  • Confirm OLLAMA_BASE_URL is correct for your OS.
  • Restart stack: docker compose down && docker compose up --build -d.

Recommended Models for Audify

  • qwen3:1.7b for fast local testing.
  • llama3 for broader language quality.
  • mistral as another lightweight option.

Choose based on latency, quality, and available hardware.