AI & Your Data · Lirenota

Understanding runs on your iPhone unless you choose a cloud provider. Nothing goes to the cloud without your choice.

How reports work

Speech is turned into text, then Lirenota interprets your words and checks results before saving. Clear items may save automatically; unclear items can need your input. Review all entries before relying on them.

Choose where processing happens

On device: Apple on-device features, where supported, and Lire Forte or Lire Breve process reports on your iPhone. Speech and text routes are separate choices.
Cloud: if you select a cloud route and allow it, report text or audio is sent to the selected provider: OpenRouter, OpenAI or Anthropic. OpenRouter may route your request to your selected model provider. Lirenota does not silently switch to cloud processing.

Permission and data sent

Cloud processing is opt-in. The app asks for permission when needed and remembers your choice. You can change it in Settings under Allow cloud processing; withdrawal stops future cloud processing, not processing already completed.

A request can include your report text or recorded spoken turn, instructions and relevant context needed to interpret it. Attached files are not sent to the understanding provider. Do not include information you do not want the selected provider to receive.

Provider terms and keys

Providers process and may retain requests and responses under their own terms and privacy policies, which may change. Review those terms and account settings before consenting; contact the provider about its retention or deletion. Cloud requests use your own API key, stored in the iPhone Keychain. Providers may charge you directly.

Lire and model downloads

Lire Forte and Lire Breve run on your iPhone and do not learn from your reports. Downloading them contacts Hugging Face using the app's read-only repository key. Hugging Face receives your IP address and normal device or network metadata, but no report or financial data. You do not need to enter a token.

Accuracy

AI and speech recognition can make mistakes, especially with noise, mixed languages or unusual names. Check saved items and corrections before relying on them.

Acknowledgements

Lirenota is built with open-source software and models. Thank you to the people who make them.

Lire models

Lire Forte and Lire Breve are Lirenota fine-tunes of Qwen3.5-0.8B, a base model by the Qwen team. They are modified versions trained on Lirenota internal data only.

Qwen3.5-0.8B (base model): Qwen team. Apache License 2.0.

On-device runtime

swift-transformers (tokenizers): Hugging Face. Apache License 2.0.
swift-huggingface: Hugging Face. Apache License 2.0.
swift-jinja: Hugging Face. Apache License 2.0.

Apple's open-source packages

swift-collections: Apple. Apache License 2.0.
swift-numerics: Apple. Apache License 2.0.
swift-crypto: The SwiftCrypto Project. Apache License 2.0. Copyright 2019 The SwiftCrypto Project.
swift-asn1: The SwiftASN1 Project. Apache License 2.0. Copyright 2022 The SwiftASN1 Project.

Voice detection and turn models

Silero VAD Core ML conversion and Swift wrapper: Pæan. MIT License. Copyright (c) 2026 Pæan.
Smart Turn v3.2 model, from Pipecat: Daily. BSD 2-Clause License. Copyright (c) 2024-2025, Daily.
Smart Turn v3.2 Core ML build: aufklarer/Smart-Turn-v3.2-CoreML, under the upstream Pipecat license.

Apple frameworks

Lirenota is built with Apple's SwiftUI, Speech, AVFoundation, Core ML, WidgetKit and related frameworks, used under Apple's developer terms.

License texts

Full license texts are available on request at support@lirenota.tech.