- My role
- Solo (product, design, build)
- Timeline
- May to Sep 2026
- Platform
- iOS 26
- Status
- Free on the App Store
- running cost
- $0
- servers
- 0
- on-device transcription
- 100%
Overview
I go to a lot of talks and rarely kept much of what was said. I wanted a simple tool to record what I heard and summarise it when I needed to. It had to be simple and close to free to run, otherwise I’d have paid for something already on the market. I took it from first idea to a public, free iOS app, and made every product, design and technical call along the way.
Problems
Talks slip away
Some talks are too technical to follow in the moment. Others get interrupted by someone next to you starting a conversation. Either way, the part you missed is gone.
Existing tools don’t fit occasional use
Transcription apps are built around monthly subscriptions. That works for daily meetings, not for someone who goes to a conference once a month.
Framing
Build the smallest thing that reliably captures a talk, at close to zero cost. Recording and transcription are the product. Summaries are optional. Anything that adds running cost or complexity has to earn its place.
The work
1. Start on the web to move fast
I built the first version as a web app because it was faster to develop and test. It used cloud services for transcription and summaries. Once the core flow held up in real use, I migrated it to iOS, where recording naturally belongs.
2. Move transcription onto the phone
The cloud approach kept failing in the field. Recordings were lost when there was no signal, and long sessions produced files too heavy for the APIs to process. Moving transcription onto the device solved all of it at once: it works offline, has no file size limit, and costs nothing per recording. Recording also keeps running with the screen off, so the phone can stay in your pocket. The tradeoff is that speakers aren’t identified. For recalling what was said at a talk, that’s acceptable.
3. Make recordings impossible to lose
If a recording fails, nothing else matters. Text is saved as it’s spoken, and a session interrupted midway is recovered instead of discarded.
4. Keep the recording screen focused
Recording takes over the whole screen so the live transcript stays readable. Discarding a recording asks for confirmation instead of deleting in one tap.
5. Build the system before the screens
Every style was tokenized from day one, and every component exists as a reusable component in the code, mirrored in Figma. Colours, states and forms follow a small set of rules, with every colour passing WCAG AA. Changing the app means changing a token or a component once, not hunting through screens.
6. Remove the server
Data now lives in the user’s own iCloud. There’s no backend to maintain, pay for, or secure. A side effect is that nothing a user records passes through anything I operate, which makes the app suitable for more sensitive uses than conference talks.
7. Release it for free
A subscription made no sense for the same reason I didn’t want to pay for one. I designed a full paid model before deciding against it: with no running costs, there was nothing honest to charge for. The only costly feature is the AI summary, and it’s optional. Users who want it plug in their own Anthropic API key, and summaries go directly from the phone to Anthropic.
There’s no tracking or analytics. Instead, users can send anonymous feedback from inside the app, so if people do pick it up, I still hear about bugs and feature requests. I chose this over opening an email draft, which would have exposed the user’s own address.
