End User License Agreement

DRAFT — pending legal review (S-Legal-3). This outline shows the intended shape of the EULA; the reviewed text replaces it before commercial launch.

The final EULA will cover, in plain language:

  • License grant — a perpetual, non-exclusive license to install and use the app on the device slots your purchase covers (see thelicensing docs); the trial grants the same, time-limited, reverting to read-only.
  • What you own — your notes and data, entirely; they live on your device and never become ours.
  • What we own — the application software. The license is a right to use, not a transfer of the software's ownership.
  • Third-party components — AI models downloaded by the app are third-party works licensed separately by their creators (see the model schedule below — the only place the exact models are named; the product itself speaks incapability tiers); open-source libraries are acknowledged in-app (Settings → About).
  • Restrictions — no resale or sublicensing of the app or license codes, no circumvention of the license system; no restrictions on what you write or export.
  • Updates & major versions — updates are provided for the purchased major version per the support policy; major upgrades are separate purchases.
  • Warranty & liability — provided as-is within the limits of applicable law; local storage means backups are your responsibility (the app provides tools).
  • Termination — what happens on refund or license revocation (app reverts to read-only; your data remains exportable).

Appendix — Third-party model schedule (current release)

This schedule is the single disclosure of the exact AI models the app downloads and runs. Everywhere else — in the app and on this site — models are described only by capability tier, parameter class, and quantization. Tier pins change per release; this schedule changes with them.

SurfaceModelSource repositoryLicense
Lite tier · caption companion on larger Apple Silicon tiersQwen 3.5 4B (Q4_K_M)huggingface.co/unsloth/Qwen3.5-4B-GGUFApache 2.0
Standard tier · Lite quality option (8 GB GPUs, gated) · Moderate tier on 16 GB GPUs (Q6_K)Qwen 3.5 9B (Q4_K_M / Q6_K)huggingface.co/unsloth/Qwen3.5-9B-GGUFApache 2.0
Moderate tier on Apple Silicon · Supreme tier on 24 GB GPUsGemma 4 26B-A4B, mixture-of-experts (MXFP4)huggingface.co/unsloth/gemma-4-26B-A4B-it-GGUFApache 2.0
Supreme tier on Apple Silicon · Ultra quality option on 32 GB GPUsQwen 3.6 27B, dense (Q4_K_M)huggingface.co/unsloth/Qwen3.6-27B-GGUFApache 2.0
Ultra tier, defaultQwen 3.6 35B-A3B, mixture-of-experts (MXFP4)huggingface.co/unsloth/Qwen3.6-35B-A3B-GGUFApache 2.0
Ultra quality option (Apple Silicon 48 GB+)Gemma 4 31B, dense (Q4_K_M)huggingface.co/unsloth/gemma-4-31B-it-GGUFApache 2.0
Semantic search (every install)paraphrase-multilingual-MiniLM-L12-v2huggingface.co/sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2Apache 2.0
Voice — language identification (every install)OpenAI Whisper basehuggingface.co/openai/whisper-baseMIT
Voice — speech-to-text engine (0.6B below 24 GB / 16 GB VRAM; 1.7B above)Qwen3-ASR 0.6B / 1.7B (Q8_0)huggingface.co/ggml-org/Qwen3-ASR-0.6B-GGUF · huggingface.co/ggml-org/Qwen3-ASR-1.7B-GGUFApache 2.0
Voice — Tamil / Telugu / Kannada packs (on-demand per language)Whisper-medium fine-tunes by vasista22 — Speech Lab, IIT Madras (supported by the Bhashini programme, MeitY, Govt. of India)huggingface.co/vasista22/whisper-tamil-medium · whisper-telugu-medium · whisper-kannada-mediumApache 2.0, attribution
Voice — Malayalam pack (on-demand)Whisper-medium fine-tune by thennal (independent contributor)huggingface.co/thennal/whisper-medium-mlApache 2.0, attribution
Voice — fallback transcriptionOpenAI Whisper smallhuggingface.co/openai/whisper-smallMIT
Image captions (every install; the floor runs it CPU-only in idle batches)Moondream 2huggingface.co/vikhyatk/moondream2Apache 2.0

Model and company names identify their owners' work and belong to their respective owners; Smritsavant is not affiliated with or endorsed by them. Weights download from the official repositories above directly to your device.

Questions meanwhile: [email protected].