Skip to content

Get started with distil labs

distil labs is a developer platform where engineers train and deploy small models that match frontier accuracy at 80% lower cost.

Start with a coding agent

Section titled “Start with a coding agent”
claude "Let's get started with distillabs.ai/onboarding.md"

Runs in your terminal.

Your agent installs the CLI, creates your account, and trains your first model with you. It stops for your go-ahead at four points, so nothing runs that you have not seen. The whole flow takes 30 to 45 minutes, and jobs run for 20 to 35 of those minutes.

Start with the CLI

Section titled “Start with the CLI”
curl -fsSL https://cli-assets.distillabs.ai/install.sh | sh
distil signup

distil signup opens your browser and hands the session back, so you finish signed in. If you already have an account, run distil auth instead. You can also sign up at app.distillabs.ai/sign-up.

The default path starts from the model you run in production today. An inference endpoint goes in front of it, records the traffic, and that traffic becomes the training data. The build overview covers each step:

distil inference-endpoint create --name support --fallback-model "openai/gpt-4.1-mini"
distil api-keys create support-prod
distil inference-endpoint link-api-key <unique-endpoint-name> support-prod
# point your application at https://inference.distillabs.ai/v1 with that key and name,
# then let it collect

distil inference-endpoint download-traces --all <unique-endpoint-name>
# convert the records, then:
distil traces upload --data ./traces-input
distil seed-dataset create-from-traces <traces-id>
# Seed dataset created with ID: <seed-dataset-id>

distil training-dataset create-from-seed-dataset <seed-dataset-id>
distil slm create-from-training-dataset <training-dataset-id>
distil deployment create-from-slm <slm-id>

distil inference-endpoint create --name support-slm --fallback-model "openai/gpt-4.1-mini" \
  --primary-url <deployment-url> --primary-api-key <deployment-api-key>
# move your application to the new name; it records again, so the loop repeats

If you already have a labelled dataset, distil seed-dataset create --data ./my-data-dir replaces the first block, and the rest is the same. Your trained model is also ready for hosted inference on its own, or to run locally.

Next steps

Section titled “Next steps”