Quickstart

1. Create a model

Sign in and go to New model. Drop in links and documents: your website, terms of service, policies, PDFs, even a competitor. PDF text stays tied to page numbers, and scanned pages are flagged for review instead of silently treated as empty. We research the evidence and draft your values, matched to a base model. Or fill everything out from scratch.

2. Describe what you stand for

The spec is plain language: how you talk, the values you hold, phrases you would never use, phrases you always want, and when the model should hand off to a human. Add a handful of demonstration examples, each one a message your business faces and the response you wish it always gave.

3. Confirm the base model

We match a base model to your sources: Gemma 4 26B for deterministic, scripted support, or Gemma 4 31B when the policy needs broader judgment. Both launch checkpoints run on validated Modal serving deployments. Bring another compatible Hugging Face checkpoint to the concierge call. Pricing is on usage and billing.

4. Ship it

Submit, and the model moves through queued, training, and evaluating (policy attestation against your values). When it reaches ready, its dashboard page shows the endpoint details, the model id, and the hash-verified attestation report.

5. Call it

The endpoint is OpenAI-compatible. Create an API key under API, then substitute your model id:

curl https://api.opensteering.com/v1/chat/completions \
  -H "Authorization: Bearer $OPENSTEERING_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "{model-id}",
    "messages": [
      {"role": "user", "content": "A customer wants a refund after 45 days. What do we say?"}
    ]
  }'

That is the whole setup. For SDK examples and streaming, continue to calling your model.