Disclosure
Last updated 6 October 2026This page says plainly how Jelly uses artificial intelligence, what it sends where, and where it falls short. It is not a legal document. It exists because a product that reads your apps owes you a clear account of itself.
Jelly is built on AI models we do not own
Jelly's reasoning is done by large language models from OpenAI, and its speech and transcription by Deepgram. We do not train our own models on your data. We send excerpts of your content to those providers, through our own service, using our credentials.
What is automated
- Reading text in the apps you choose and turning it into structured memory: people, commitments, meetings, decisions.
- Deciding a commitment exists, and deciding a later message settles it.
- Writing meeting notes, summaries, drafted replies and answers.
- Judging relevance: whether a message is addressed to you, whether two things are the same thing.
None of it sends anything on your behalf. Jelly drafts; you send.
Where it is wrong
We would rather say this here than have you discover it:
- It misreads intent, especially across languages and in mixed-language chat. A phrase that means "that works" can be read as "will that work?".
- It invents nothing knowingly: answers are grounded in what it has actually seen. But a model can still state something confidently that is not in the source.
- It misses things. A conversation it could not read is a conversation it does not know about.
- It closes a loop on evidence, and that evidence can be the wrong message. Every closure keeps the message it closed on, so you can check, and you can reopen it.
What passes through our service
Our service forwards your requests to the model providers. We do not write that content to disk, log it, or store it; we store only usage metadata. The Privacy Policy sets this out in full.
Training
Your content is not used to train any model, ours or our providers'. Data sent to the OpenAI API is not used to train or improve OpenAI's models, and every request we send to Deepgram is opted out of its model improvement program.
Recording and consent
Jelly can listen to a meeting and write it up. It does not join the call or appear in the participant list, which means the other people in the call will not be told unless you tell them. Laws on recording and note-taking differ by country and by state, and some require everyone's consent. Complying with them is your responsibility.
Beta
Jelly is in private beta. Expect defects, expect change, and do not rely on it as a system of record.