CrossPawn Prayer Voice

Draft notice: The site operator should review this notice and the hosting provider's practices before public launch.

Terms of use

CrossPawn is a research and reading aid, not a substitute for professional advice. Results and interpretations may be incomplete or incorrect; check them against the cited source text. Respect the rights and licenses of material you import or use, and do not use the site to harm others or interfere with its operation.

The existing "Mini LLM" button produces rule-based retrieval and synthesis in Search Results. When enabled, the language model adds a separate interpretation. Its instructions and simple checks against future predictions are not guarantees; outputs may be inaccurate or include source material or annotations. Review them critically.

Privacy and device resources

Research, imported files, and questions are processed in the browser. Manual and community annotations are temporary unless you choose Save Annotations Locally; saved notes remain in this browser on this device until browser data is removed or overwritten. Anyone with access to this browser profile may be able to read them. If you explicitly publish a global annotation, its text and chosen name are sent to the server and made public; do not include private information.

The app does not use your device for cryptocurrency mining or include analytics. Questions and notes are not collected unless you explicitly publish an annotation. The site host, model-file host, and network providers may nevertheless record connection information, such as IP addresses and requests for site, library, and model files, under their own practices. Your browser or operating system supplies speech playback and may handle it according to its own privacy settings.

If you enable the optional model, model files are downloaded to this device and may be cached by the browser. The model runs in your browser and uses its CPU and memory; it does not reserve a fixed amount or enforce operating-system core limits. Global annotations use separate Supabase storage for reader and AI submissions. Hosting providers may process network information for rate limiting and logs. A basic arithmetic CAPTCHA limits casual abuse but does not prevent automated spam; public posts are not moderated before publication.

Enable only if you agree to use this device's CPU, memory, storage, and bandwidth for local AI. The model is downloaded from the configured site or model-file host and may be cached. Questions and notes are processed locally, not sent to that host. Model files are publicly hosted by the site operator; the host can log file requests. Stop AI to release the worker. Notes are included only when you explicitly request an annotation summary, history search, or chapter interpretation with annotations.

The memory tiers select different models, not reserved RAM or a maximum; actual memory can be higher. Core choices are worker-thread targets, not CPU limits, and may fall back to one thread. The 4 GB tier can take several minutes per response and may exhaust available memory; use a smaller model or stop AI if your device struggles. Token progress appears after generation begins; answers appear after checks. Chapter viewing and deep reports review all loaded chapter text in separate sections, which may take a while. Other actions use selected excerpts. The AI may interpret beyond the supplied text and can make mistakes, misidentify sources, or ignore instructions; verify its claims. Basic quotation/reference checks warn about possible mismatches but do not verify accuracy. The no-prediction keyword checks are limited and may withhold valid interpretations.