docker-model

Docker Model Runner Skill

Docker Model Runner (DMR) manages and serves AI models through Docker Desktop or Docker Engine, exposing OpenAI-compatible APIs. Models are pulled as OCI artifacts from Docker Hub (ai/ namespace), any OCI registry, or Hugging Face, and stored locally. For Drupal work it provides a free, local, keyless backend for the AI module ecosystem during development.

---

Enabling

• Docker Desktop: Settings → enable Docker Model Runner (Beta features).
• Docker Engine (Linux): supported without Desktop; models are served on the host. GPU support: NVIDIA (CUDA), AMD (ROCm), Vulkan; Apple Silicon on macOS; CPU everywhere.

---

Core CLI

docker model status                      # is the runner active?
docker model pull ai/smollm2             # fetch a model (Docker Hub ai/ namespace)
docker model pull hf.co/bartowski/Llama-3.2-1B-Instruct-GGUF  # from Hugging Face
docker model list                        # local models
docker model run ai/smollm2 "Hello"      # one-shot prompt
docker model run ai/smollm2              # interactive chat (exit with /bye)
docker model configure --context-size 8192 ai/smollm2   # adjust context window
docker model inspect ai/smollm2          # model metadata
docker model logs                        # runner logs
docker model rm ai/smollm2               # delete local model

Run docker model --help for the full, current command list — the CLI is still evolving.

---

OpenAI-compatible API

Endpoint	Method
/engines/v1/models	GET
/engines/v1/chat/completions	POST
/engines/v1/completions	POST
/engines/v1/embeddings	POST

Base URLs:

• From the host: http://localhost:12434 (default TCP port)
• From containers (Docker Desktop): http://model-runner.docker.internal
• From containers (Docker Engine): http://172.17.0.1:12434

curl http://localhost:12434/engines/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model": "ai/smollm2", "messages": [{"role": "user", "content": "Hi"}]}'