Models we use
Which AI models LoreClerk runs, for what, and what the providers may do with your text.
| Model | Made by | Reached through | Used for |
|---|---|---|---|
| Claude Haiku 5.5 | Anthropic | OpenRouter, on Google Vertex AI or Amazon Bedrock | Continuity checks, story-note extraction, summaries, and fix suggestions. |
| DeepSeek V4.1 Flash | DeepSeek | OpenRouter, on independent zero-retention hosts | Drafting chapters and briefs, rewriting selections, translation drafts, the new-book wizard. |
| Your own key | The provider you choose | Your account | Writer and above can connect their own provider; those tasks go there instead, under your account and its terms. |
Every request goes through OpenRouter with zero data retention: it only reaches hosts that don’t store prompts or replies, so they can’t train on them. OpenRouter itself doesn’t keep the text after answering; it keeps the time, model and token counts for billing. Its one stated exception is screening for child sexual abuse material, which may be kept for review and reporting where the law requires.
The model makers, Anthropic and DeepSeek, don’t receive your text: the models run on the hosts named above. Repeated instructions may sit in a host’s short-lived memory cache to save cost; nothing is written to storage.
We don’t generate images, audio or video. Every model works on text you wrote or asked for.
Instructions you type for AI features are checked for safety first; see the Acceptable use policy.
When we change a model, this page changes first.