# Sweet Pal

> Sweet Pal is an AI companion app for iPhone for people living with type 1 diabetes: you tell it about meals, movement and how you feel, it organises the numbers and learns how you work. It does not deliver insulin or control devices, and nothing it shows is medical advice: dosing decisions stay with the person and their diabetes team. The app currently speaks Italian. Site in Italian (default) and English (/en).

## Product
- [Home (IT)](https://sweetpal.ai/): what Sweet Pal is
- [Home (EN)](https://sweetpal.ai/en): what Sweet Pal is, in English
- [For people with T1D](https://sweetpal.ai/en/for/t1d-patients)
- [For clinicians](https://sweetpal.ai/en/for/t1d-docs)
- [For tech fans](https://sweetpal.ai/en/for/tech-fans)
- [Contact](https://sweetpal.ai/en/contact)

## Newsroom (official news)
- [Sweet Pal closes its beta: App Store release is near](https://sweetpal.ai/en/newsroom/sweet-pal-beta-ends.md): Sweet Pal, the AI companion for people living with type 1 diabetes, has closed its beta phase and is getting ready for release on the App Store for iPhone.
- [Sweet Pal chiude la beta: il rilascio su App Store è vicino](https://sweetpal.ai/newsroom/fine-beta-sweet-pal.md): Sweet Pal, il compagno con l'intelligenza artificiale per chi vive con il diabete tipo 1, chiude la fase beta e si prepara al rilascio su App Store per iPhone.

## Blog — English
- [AI for type 1 diabetes: what it must do, and never do](https://sweetpal.ai/en/blog/ai-type-1-diabetes-safety.md): An AI assistant for type 1 diabetes can observe, show and explain, but must never decide a dose. The safety rules we built into Sweet Pal, and why. (T1D)
- [What an AI chat turn really costs: tokens, cache, models](https://sweetpal.ai/en/blog/ai-chat-cost-per-turn.md): AI chat cost lives in the tokens a model reads, not the ones it writes. Multiple rounds, prompt caching, one model per role and the real cost per subscriber. (AI)
- [Session handoffs: keeping a coding agent on track for weeks](https://sweetpal.ai/en/blog/coding-agent-handoff-sessions.md): A coding agent starts from zero every session. How a handoff file, read first and rewritten last, keeps it on track on the same product for weeks. (Vibe coding)
- [LLM evaluation before release: sentences, versions, first token](https://sweetpal.ai/en/blog/llm-evaluation-before-release.md): How we do LLM evaluation before release: reference sentences against the real model, versions switched off but never deleted, and time to first token. (AI)
- [Vibe coding with rules: why a CLAUDE.md beats a clever prompt](https://sweetpal.ai/en/blog/vibe-coding-project-rules.md): Vibe coding on a real product only holds up when the rules live in files the agent rereads every session. Ours, one by one, and the reason behind each. (Vibe coding)
- [Decide first, speak later: how J2LM made our chat fast](https://sweetpal.ai/en/blog/j2lm-decide-first-speak-later.md): We took the deciding away from the LLM: JOLs classify in parallel, a script writes the brief, the model only speaks. Sweet Pal now replies in about 1.5 s. (AI)

## Blog — Italiano
- [AI e diabete di tipo 1: cosa deve fare e cosa mai](https://sweetpal.ai/blog/ai-diabete-tipo-1-sicurezza.md): Un assistente AI per il diabete di tipo 1 osserva, mostra e spiega, ma non decide dosi. Le regole di sicurezza che abbiamo scritto in Sweet Pal, e perché. (T1D)
- [Quanto costa un turno di chat AI: token, cache e modelli](https://sweetpal.ai/blog/costo-chat-ai-per-turno.md): Il costo di una chat AI sta nei token letti, non in quelli scritti. Giri multipli, prompt caching, un modello per ruolo e il costo vero per abbonato. (AI)
- [Handoff fra sessioni: un agente di coding che non perde il filo](https://sweetpal.ai/blog/agente-coding-handoff-sessioni.md): Un agente di coding riparte da zero a ogni sessione. Come un file di handoff, letto per primo e riscritto alla fine, lo tiene sul pezzo per settimane. (Vibe coding)
- [Valutare un LLM prima di accenderlo: frasi, versioni, primo token](https://sweetpal.ai/blog/valutare-llm-prima-del-rilascio.md): Come valutare un LLM prima del rilascio: frasi di riferimento contro il modello vero, versioni spente e mai cancellate, e il tempo del primo token. (AI)
- [Vibe coding con le regole: perché un CLAUDE.md batte il prompt](https://sweetpal.ai/blog/vibe-coding-regole-progetto.md): Il vibe coding su un prodotto vero regge solo se le regole stanno in file che l'agente rilegge a ogni sessione. Le nostre, una per una, e il perché di ognuna. (Vibe coding)
- [Prima decidere, poi parlare: come J2LM ha accelerato la chat](https://sweetpal.ai/blog/j2lm-prima-decidere-poi-parlare.md): Abbiamo tolto le decisioni al modello linguistico: i JOL classificano in parallelo, un copione scrive le istruzioni, il modello parla. Risposte in circa 1,5 s. (AI)

## Optional
- [Blog RSS (EN)](https://sweetpal.ai/en/blog/feed.xml)
- [Blog RSS (IT)](https://sweetpal.ai/blog/feed.xml)
- [Newsroom RSS (EN)](https://sweetpal.ai/en/newsroom/feed.xml)
- [Privacy](https://sweetpal.ai/en/privacy-policy)
- Press contact: news@sweetpal.ai