The question I get most often about Memglow isn't about features. It's some version of this: I just recorded a voice note about my friend's health scare. Where did that go?
It's the right question to ask. What you tell Memglow about people is sensitive by nature, and I think you deserve a plain answer rather than a link to a policy document. So here's the honest walkthrough of what happens to a voice note, step by step, and the decisions we made at each one.
Your phone. You record. The audio leaves over an encrypted connection.
Transcription. OpenAI's speech API turns your words into text. The audio is then deleted.
Extraction. Anthropic's Claude finds the facts worth keeping and scores its confidence.
Your review. Nothing is saved until you confirm it.
Storage. Text only, encrypted at rest, readable by your account alone.
Step one: the recording leaves your phone
Let's start with the part some apps would bury: Memglow's AI does not run on your phone. When you finish recording, the audio travels over an encrypted connection to our servers, and from there to the AI providers that do the heavy lifting. OpenAI's speech API turns your words into text, and Anthropic's Claude reads that text to pull out the facts. Your content is not used to train their models, and exactly how each provider handles it is spelled out in their policies, which we link from our Privacy Policy.
We built this pipeline to be tight at every step: encryption in transit, encryption at rest, no AI credentials anywhere near your device, and providers bound by their published policies. Security isn't a checkbox we tick before a release. It's the first question we ask when we design anything, and it always will be.
Step two: transcription
Your audio comes back from the transcription step as plain text. This is the moment your spoken words become the written note you see in the app.
And here's the decision that matters: once transcription succeeds, the audio has done its job. On the Core plan, the recording is deleted right after. Not archived, not retained for quality purposes, deleted. Pro users can choose to keep recordings for 60 days if they want to listen back, but that's an opt in, with a ceiling, and a clock on it.
Step three: extraction
The transcript then goes to Claude with one job: find the facts worth keeping. Her partner's name is Theo. The studio is in Bushwick. Her parents visit next month. Each fact comes back with a confidence score, and anything that touches health, relationships, or similar territory gets flagged as sensitive automatically.
Then everything stops and waits for you. Nothing is saved to a profile until you've reviewed the list, fixed what's wrong, and confirmed what's right. The AI proposes; you decide.
What's stored, and who can see it
What remains after all this is text: your note, the facts you approved, and the summary the app writes from them. That data is encrypted in transit and at rest, and it's walled off at the database level so that your rows are readable by your account and no one else's. There's no social layer, no shared graph, and we don't sell data. The business model is the subscription, full stop.
Facts flagged sensitive get an extra layer in the app itself: they render blurred until you tap to reveal them, so a glance over your shoulder shows dots instead of details. You can also put the entire app behind Face ID.
The exit door
One more decision worth knowing about. If you delete your account, everything goes: contacts, memories, facts, any retained audio, the account itself. It's a full cascade, not a soft deactivation. Your data shouldn't be a hostage to a subscription you no longer want.
That's the pipeline. If you want the formal version with all the specifics, it's in our Privacy Policy. And if you have a question this post didn't answer, email support@memglow.ai and ask. The whole point of building a memory app for personal relationships is that trust has to come first.