Evaluation-Dataset Preference
By default, your prompts (after PII scrubbing) may be included in our private, internal datasets used to benchmark LLM quality. They are never sold, published, or shared. You can opt out below — your access is unaffected either way.
Loading…
Your current status
What evaluation datasets are
- By default, your prompts and responses (after PII scrubbing) may be included in our private, internal datasets used to benchmark LLM quality.
- They are never sold, licensed, published, or shared. We may publish aggregate benchmark results and open-source the eval harness, but never the underlying dataset of your content.
- IPs, User-Agent strings, HTTP headers, and account identifiers are never included — only post-scrub content. The scrubber (Microsoft Presidio + spaCy NER + custom recognizers) is best-effort, not a guarantee of perfect de-identification. See Privacy Policy §3a.
What stays the same either way
- You keep your account and API key, and all models stay available — opting out never changes your access.
- Your requests are still logged and used for service operation (security, abuse detection, debugging). Operational logging is independent of this preference.
Lawful basis & your right to object
We include content in evaluation datasets on a legitimate-interest basis (GDPR Art 6(1)(f)) — measuring and improving the quality of the inference we provide. You may object at any time (Art 21) by opting out above. See Privacy Policy §2b.
Model Training & Premium Models
Separately, you can opt in to let us use your post-scrub content to train and fine-tune our models. Opting in unlocks premium models. This is entirely optional — the standard models work without it.
Your training status
Before you opt in — what training means
- The exchange: opting in lets us use your post-scrub prompts and responses to train/fine-tune models, and in return you unlock premium (newer/larger) models. The standard tier stays free and fully usable either way.
- Consent is the lawful basis (GDPR Art 6(1)(a)). You can withdraw any time — that stops future training use and revokes premium access.
- Withdrawal is not retroactive for already-trained models. Content already incorporated into trained model weights cannot be removed from those weights — training is irreversible at the weight level. Deleting your account removes your logs and excludes you from future training, but cannot un-train an existing model. See Privacy Policy §2c.
- We still never sell, license, or distribute your data — or any derivative of it. Training use is strictly internal.