If you find this add-on useful, please star it on GitHub — stars show appreciation and help maintainers know their work matters.

DDEV Assistant Ollama

Overview

This add-on installs Ollama inside the web container of your DDEV project, so local models are available to your code with no extra setup.

The ollama command works in ddev ssh and ddev exec, and the Ollama API answers on http://localhost:11434 inside the web container. Models are downloaded once per machine and shared by every project that uses this add-on.

Requirements

DDEV >= v1.24.10

Installation

ddev add-on get brookemahoney/ddev-assistant-ollama
ddev restart

After installation, make sure to commit the .ddev directory to version control.

Usage

Command	Description
`ddev exec ollama run <model>`	Run a model, downloading it first if needed
`ddev exec ollama list`	List downloaded models
`ddev exec ddev-ollama-start`	Start the server, if it is not already running
`ddev describe`	View the Ollama endpoint in the service table

From inside the web container, the API is at http://localhost:11434, or http://web:11434 from another container in the same project:

ddev exec ollama run llama3.2
ddev exec curl -s http://localhost:11434/api/version

Configuration

The add-on sets two environment variables, and only as defaults, so a project can override either of them with web_environment in .ddev/config.local.yaml:

Variable	Default	Description
`OLLAMA_HOST`	`0.0.0.0:11434`	Where the server listens. Point it at a non-local address, such as `host.docker.internal:11434`, to use an Ollama that already runs on the host; the add-on then does not start a server of its own.
`OLLAMA_MODELS`	`/mnt/ddev-global-cache/assistant-ollama/models`	Where models are stored. The DDEV global cache is per machine, so a model is downloaded once for all projects.

# .ddev/config.local.yaml
web_environment:
  - OLLAMA_HOST=127.0.0.1:11435

Server logs are in /tmp/ollama.log inside the web container:

ddev exec tail -f /tmp/ollama.log

Reaching Ollama from the host

The port is not published by default, because the server is meant for the web container. To reach it from the host as well, publish it on the web service:

# .ddev/docker-compose.web-ports.yaml
services:
  web:
    ports:
      - "11434"

ddev describe then shows the host port to use.

How it works

Installs Ollama into the web image at /usr/local/bin/ollama, with its runtime in /usr/local/lib/ollama, so it is on $PATH in every shell type and survives ddev restart without a per-start copy
Starts the server on start with a post-start hook that waits until the API answers, so Ollama is usable as soon as ddev start returns. The hook is idempotent, and it fails loudly rather than leaving you with no server
Keeps models out of the image and out of the project, in the DDEV global cache, which is shared across projects and survives ddev restart, ddev rm and ddev add-on remove

Notes

The CPU-only Ollama build is installed, so no GPU is used. A web container has no GPU passed through to it, including on a Linux host with an NVIDIA card. Use a sidecar add-on such as tyler36/ddev-ollama if you need GPU acceleration.
Do not combine this add-on with a sidecar Ollama add-on; two Ollamas in one project will fight over the same port.
The runtime comes from the alpine/ollama image, which carries the CPU-only runtime in 68 MB rather than the ~1.5 GB of the official Linux release, most of which is GPU libraries. It tracks Ollama releases a week or two behind; ddev exec ollama --version shows the installed version.
ddev add-on remove brookemahoney/ddev-assistant-ollama removes the add-on’s files, and the next ddev restart rebuilds the web image without Ollama. Downloaded models stay in the global cache; remove them with ddev exec sudo rm -rf /mnt/ddev-global-cache/assistant-ollama.

Credits

Contributed and maintained by @brookemahoney

If you find this add-on useful, please star it on GitHub — stars show appreciation and help maintainers know their work matters.