CLI Quickstart

Deploy a trained model artifact in one command using the deployment CLI.

---

Install CLI extras

uv sync --extra cli   # or: pip install -e ".[cli]"

---

Set your API key

export GROQ_API_KEY=<your-groq-api-key>

The CLI uses Groq to generate load() and predict() method bodies for your model.

---

Deploy a model

inference-engine deploy ./my_model.pkl

The CLI will: 1. Inspect the artifact (framework, input shape, class labels) 2. Prompt for name, version, device, and routing strategy 3. Generate serving code via LLM 4. Validate the pipeline against a sample input 5. Show a preview and write files on confirmation

---

Non-interactive deploy (CI)

inference-engine deploy ./my_model.pkl \
  --name my_model --version v1 \
  --device cpu --routing static \
  --sample-input '{"features": [1.0, 2.0, 3.0]}'

---

Fix a broken pipeline

inference-engine fix models/my_model/v1/

Reads the existing definition.py, validates it, and sends failures to the LLM for repair.

---

See CLI Reference for full documentation.