A work conversation ends, but the people stay at the table. The discussion moves from pricing and deadlines to something personal. You may want to keep a clear record of the agreement and leave the next part out of your history altogether.
The microphone cannot make that choice just because the meeting is over. You need a way to stop it, and the people with you need to know that they can ask.
This is the design question behind Draki: how do you make conversations easier to remember while leaving room for conversations that should simply end?
You cannot choose every useful sentence in advance
It is tempting to say that a memory app should keep only the important bits. The difficulty is knowing what will matter later.
A customer describes how their team handles a purchase. You remember the budget discussion. Three weeks later, you discover that the approval process was the detail you needed. A summary centered on price may have left it out.
There is a reason to keep enough of a chosen conversation to revisit its meaning. A promise can have a condition. A decision can follow an objection. A person's view can change halfway through the discussion. If you preserve only the final sentence, you may lose the explanation that makes it useful.
That argues for a transcript you can check. It does not settle whether every interaction belongs in the app. Choosing to record a customer discussion and choosing to record the rest of your day are separate decisions.
Finding a detail should not require rereading the day
More storage does not make a useful answer by itself. You can have a complete transcript and still spend twenty minutes hunting for the right passage.
What you want back is usually smaller than what you saved: the reason a deal stalled, the name of the person who offered an introduction, the change a customer asked you to make. The date matters, too. A preference from last month may have been replaced yesterday.
Draki's conversation memory is built around returning to that context. The current app sends audio through Draki to OpenAI over Wi-Fi after your permission, with text stored for cloud summaries and search. Draki's custom MCP connection is optional and read-only. It works with AI assistants that support remote MCP, subject to their account requirements and settings. An authorized assistant can search shared text and context by topic, time or confirmed person and retrieve transcript passages. That makes it possible to ask about a particular relationship without reading every meeting again.
The source remains useful. If the answer changes what you plan to do, you should be able to check the words behind it. A short response and a fuller record can serve different purposes.
Keeping a conversation is a decision you can revisit
A project ends. A personal discussion no longer belongs in your working history. Someone asks about a recording involving them. The reason you once kept a conversation may no longer apply.
This is where "your data" has to mean something practical. You need to know how to request a copy, what a removal covers and whether another assistant has already received information from the conversation.
The Draki app works without a bracelet. The bracelet makes capture convenient to keep available throughout the day, while leaving you free to turn recording on or off whenever you want. For existing history, support can help you manage local records and handle requests about your account or information shared with an assistant. The data-control guide explains how to make the scope clear.
Stopping a microphone does not erase what it already captured. Disconnecting an assistant stops future retrieval through that connection, but does not delete information the provider received earlier.
The design ambition and the product today
We want choosing what to carry forward to be an ordinary part of using Draki. Starting capture, stopping it, checking a source and removing information should be understandable actions. A device that stays with you throughout the day still needs those boundaries.
That is a design ambition, not a claim that Draki automatically decides which facts deserve permanent memory. The current product does not promise a temporary memory layer that silently forgets everything except selected facts.
Recordings on your phone and transcripts and summaries synced to your account are separate records. Check which associated records a removal covers. The optional MCP connection shares authorized text and context through the cloud. It does not share audio or voice profiles. The current policy and storage guide explain those boundaries.
For the conversation at the table, the useful outcome is simple. You can return to the agreement when you need it. When the discussion turns personal, you can stop recording. Later, you can reconsider what you kept. Each choice belongs in a product intended to be with you every day.
FAQ
- Does an AI memory assistant need to record all day?
- No. You can use the Draki app alone, or add the bracelet to keep capture conveniently within reach throughout the day. Either way, turn recording on or off whenever you want. Wearing a bracelet does not commit you to recording every conversation.
- Does Draki automatically decide which memories to keep?
- This article describes a design view, not an automatic forgetting feature. You choose what to record. Audio, transcripts and saved contacts have different lifetimes across the phone and cloud; support can help you manage or remove them.
- Does pausing Draki delete the earlier conversation?
- No. Pausing or stopping controls further capture. Removing an existing conversation, a voice profile or copies already shared with an assistant is a separate action.
- Why keep a transcript as well as a summary?
- A summary may leave out the condition attached to a promise or the context behind a decision. The transcript gives you a way to inspect the source when a detail matters.
- How can I remove history I no longer want?
- You can delete stopped conversations in Transcripts. Support handles account deletion and can help with retained test voice data and information shared with an assistant. State the scope you want covered. Information already received by another AI provider requires that provider's own controls.
Written by Darijan Ducic
Draki for iPhone
