Chat with your notes: ask questions about any recording, entirely on your device

Most AI notetakers give you a transcript and a summary. You scroll through it. You ctrl+F for a keyword. You skim three paragraphs hoping to find the thing you half-remember being said forty minutes in.

Kuulo does something different. You can talk to your recording.

Ask it a question in plain English — "What did Sarah say about the budget?" or "Did we agree on a launch date?" — and it answers from what was actually said, with the relevant quote, attributed to the right speaker. No scrolling. No ctrl+F. No re-reading a transcript you already had to generate.

No other on-device AI notetaker does this. The few cloud tools that offer conversational AI over recordings — Otter's AI Chat, Fireflies' AskFred — send your recording to a server to process it. Kuulo's entire chat layer runs on your iPhone or Mac. The conversation between you and your notes stays on your device.

Why this matters more than it sounds

McKinsey's research on knowledge worker productivity found that employees spend nearly 20% of their workweek searching for information or tracking down colleagues who have it. A significant portion of that search is through their own meeting records: transcripts, email threads, hastily written notes from calls weeks ago.

The standard approach — scan the transcript, ctrl+F, scroll — is fine when you remember exactly what to search for. It fails in the scenarios that matter most: when you half-remember something, when you need context rather than a keyword, when someone said something in passing that you didn't flag at the time.

Chat with Notes is a different retrieval model. Instead of searching for a term, you ask a question. "What were the three objections the investor raised?" "What did we agree to deliver by Friday?" "Did James say anything about the Q3 numbers?" The AI answers from the recording — not from the internet, not from general knowledge, but from what was actually said in that specific conversation.

This is the difference between a filing cabinet and a colleague who was in the room with you.

What "on-device" actually means for AI chat

Every other tool that offers AI chat over your recordings is doing the same thing: sending your audio or transcript to a cloud server, running a large language model on that server, and returning the answer. This is why Otter AI Chat requires an internet connection. It's why Fireflies' AskFred requires a cloud account. The AI isn't running on your phone — it's running in a data centre somewhere, and your conversation content got there via an upload you may not have thought about.

Kuulo's chat runs on the Neural Engine built into Apple Silicon — the same chip architecture that Apple has used to run on-device Siri and on-device translation for several years. As Apple's own documentation confirms, the Neural Engine in current iPhone hardware (A17 Pro onwards) and M-series Macs performs trillions of operations per second — enough to run the language model required for intelligent retrieval over a recorded conversation without touching the network.

The practical implication: you can ask Kuulo about a therapy session, a board meeting, or a client call and nothing about that conversation reaches a server. For lawyers, clinicians, journalists, and anyone with confidentiality obligations, this isn't a privacy preference — it's a professional requirement.

The retrieval use cases that actually come up

"What did [person] say about [topic]?"

This is the most common query. Kuulo's diarization already knows who said what (see Speaker Diarization Explained). Chat with Notes queries that attributed structure — so "What did Marcus say about the timeline?" draws from Marcus's contributions specifically, not from everything said about timelines.

For consultants managing multiple client relationships across a week, this is the feature that changes the Monday morning catch-up. You recorded the call on Thursday. It's now Sunday night and you're preparing the client update. You ask: "What did the client say they needed by the end of the month?" and you get the exact quote, not your paraphrase of it.

"What was agreed / what are the next steps?"

The single most common information-retrieval failure in professional life: the meeting ends, people leave with different memories of what was decided. Two weeks later, an email thread starts about whether a commitment was made or not.

Kuulo's structured summary already extracts key decisions and action items. Chat allows you to probe further: "Did we agree to a fixed or variable scope?" "What was the exact wording of what we committed to?" The answer comes from the transcript — verbatim — not from the AI's best guess.

"I remember someone saying something about [topic] — what exactly did they say?"

The half-remembered remark. This happens most often in long meetings or interviews, where an important point was made early on and the rest of the conversation moved on from it. You know something was said. You can't find it in a ctrl+F search because you don't know the exact words.

Kuulo allows you to describe what you're looking for conceptually — "I think someone mentioned a number around 40% early on, what was the context?" — and the AI finds it, even if your query doesn't match the literal words used. This is semantic retrieval, not keyword search. On a device. With no upload.

Revision and research use cases

For students with a semester of recorded lectures, Chat with Notes is a revision tool. "What did the professor say about consideration in contract law?" "What was the example used to explain comparative advantage?" You're querying your actual lectures, not a textbook — the answers reflect how your specific professor explained things, which is often exactly what exam questions are written against.

For PhD researchers with dozens of qualitative interviews, it's an analysis accelerator. "Which participants mentioned job insecurity?" "What was the most common reaction to the policy change?" You're running a first-pass thematic analysis across your own transcripts, on your device, without uploading participant audio to any external service. For research ethics compliance, this matters — as covered in AI Transcription for Qualitative Researchers.

How it compares to the cloud alternatives

Otter AI Chat (Otter.ai) — Available on paid plans ($16.99/month). Sends your meeting transcripts to Otter's servers to process queries. Requires internet. Not suitable for confidential content.

AskFred (Fireflies.ai) — Conversational AI over Fireflies transcripts. Cloud-based, requires Fireflies subscription. Audio is processed and stored on Fireflies' servers. The query runs over cloud-hosted content.

Granola, Fathom, Read.ai, tl;dv, Jamie — None of these currently offer conversational AI retrieval over individual recordings. Their search is keyword-based. (Kuulo vs Granola, Kuulo vs Fathom, Kuulo vs tl;dv)

Kuulo — Conversational AI runs entirely on-device using the iPhone's Neural Engine. No internet required. No audio or transcript transmitted to process the query. Works for any recording in your archive, not just the most recent meeting. Available to all users.

The architecture gap is not a minor implementation difference. Running a language model capable of semantic retrieval over a private document requires significant compute. Until Apple Silicon made that compute available in a phone and a laptop, it simply wasn't possible to do this without a cloud server. It's possible now. Kuulo is the first on-device notetaker to build it.

The privacy case

Every conversation you've recorded carries information that was said in confidence: client commitments, medical information, investment strategy, source details, exam preparation, therapy content. When you ask an AI a question about that content, the query and the content are processed together.

With a cloud tool, that processing happens on a server that is not yours, operated by a company whose privacy policy you probably haven't read in full, subject to legal data requests you would not be told about.

With Kuulo, the processing happens on your device. The question you asked and the recording it searched are never transmitted. The answer exists only on your phone.

For the majority of cloud AI tools, their privacy policy accurately describes what happens. For Kuulo, the claim is architectural: there is no server involved in the process, so there is nothing to breach, subpoena, or share.

A note on what Chat with Notes isn't

Chat with Notes retrieves and synthesises from your own recordings. It does not connect to the internet. It does not know about things you haven't recorded. It does not replace the transcript — it queries it.

The best way to think of it: it's a researcher who was in every meeting with you, remembers everything that was said, and can answer your questions about any of it — without sharing those conversations with anyone else.

That's a new category of tool. It happens to fit in your pocket.