draft.net

faq

Straight
answers.

Common questions, honest replies.

Which model does the assistant use? +

Anthropic Claude Sonnet 4.5 by default - the best writing model widely available right now. Legacy users who prefer OpenAI can switch in settings (GPT-4o-mini).

Do you train on my writing? +

No. Your drafts, prompts, chat history, and uploads are never used to train any model. They're yours.

Can I take my data with me? +

Yes. Settings → Data → Download .zip gives you a JSON archive of everything you've created, any time. If you want to leave, account deletion is one button away, finalized by a nightly job after a 7-day grace window.

How does sharing work? +

Every generation is private by default. Toggle "Share" on an individual draft and you get a public URL like /g/abc123 that anyone with the link can read. Revoke instantly; it also suppresses SEO indexing.

Is there an API? +

Yes - personal API keys (dnk_...), Bearer auth, JSON responses, rate-limited. Details at /api. Webhooks push events to your own server for automation.

What happens when I cancel a subscription? +

You stay on your paid tier until the end of the current billing period, then auto-downgrade to Free. History stays intact; only usage quotas change.

Image generation uses which model? +

OpenAI gpt-image-2 by default (the latest in the gpt-image family, successor to DALL-E 3), with automatic fallback to DALL-E 3 if your account lacks access to the newer model.

Where are you hosted? +

AWS Lightsail in us-east-1. Postgres on RDS. Images to S3. Nothing exotic.