Skip to content

Local models (discover, pull, image)

DOTS separates three ideas:

Action Command Requires ai-server?
Inventory (read-only) ./dots models discover No
Install / pull ./dots models or ./scripts/pull_models.sh No
Compose inference server ./setup.sh --with ai-server, then dots ai … Yes

Registry and tiers: configs/models.toml in the repo and the generated models table.

Discover

Works on any machine with Python 3.11+ (same as other DOTS helpers). It scans convention paths, Hugging Face cache, Ollama, llama.cpp cache listing, Draw Things checkpoints, and—when configured—the ai-server models_dir.

./dots models discover
./dots models discover --verbose
./dots models discover --json

Managed GGUF for the Docker stack is listed separately:

dots ai models          # requires ai-server.toml

Ollama and llama.cpp

Enable runtimes with setup, then pull from the registry:

./setup.sh --with ollama,llamacpp
./dots models
# or
./scripts/pull_models.sh --provider ollama --tier auto
./scripts/pull_models.sh --provider llamacpp --tier balanced --yes

Verify with ./dots models discover.

Draw Things and img

Image generation uses Draw Things via draw-things-cli (macOS-oriented).

  1. Install the component:
./setup.sh --with drawthings
  1. Confirm the CLI:
command -v draw-things-cli
draw-things-cli models list --downloaded-only --offline
  1. Download a checkpoint (explicit opt-in; img does not pull for you):
./scripts/pull_models.sh --provider drawthings --tier balanced --yes
# or
draw-things-cli models ensure --model flux_2_klein_4b_q6p.ckpt

Example ids live in generated models.

  1. Discover or list:
./dots models discover
  1. Generate:
img -o /tmp/pic.png "An egg with legs"
img --model flux_2_klein_4b_q6p.ckpt "prompt"

See also the tools catalog and Agents → Models.

ai-server (optional)

For Open WebUI + llama-server in Docker, enable ai-server, adopt a GGUF into models_dir, set default, then start the stack. Discovery is not limited to that workflow—use ./dots models discover first, then adopt when you want a model managed for the server.

Full lifecycle: AI inference server.