Approach
Storing the audio is the easy way to build this.
Keeping every recording makes almost every feature simpler. It also turns a note taking tool into an archive of everything anybody said near your microphone. Amara AI is an attempt to build the useful half without the archive, and this page is the argument for why that trade is worth making.
The one rule
Discard the audio, keep the note
Every other decision on this site follows from that sentence, including the awkward ones.
Speech is the densest personal information most people produce. A recording carries the words, and it also carries who was there, how they sounded, what they hesitated over, who interrupted whom, and the voice itself as a biometric. A written note carries almost none of that. It carries the part you wanted.
So the question a voice product has to answer first is not how good the transcription is. It is what happens to the recording afterwards. Most products answer by keeping it, because keeping it makes search work, makes the demo better, makes re-processing possible when the model improves, and makes the retention policy a paragraph rather than a design constraint.
We are answering the other way. The recording is destroyed once the note exists. That decision costs us features, and the honest thing to do is list them rather than pretend the trade is free.
What discarding the audio costs
| Not possible | Why | What you get instead |
|---|---|---|
| Search across everything you ever said | There is no archive to search | Notes in your own system, searched by whatever already searches it |
| Re-running a better model over old meetings | The old audio no longer exists | Better notes from the day the better model arrives, and no back catalogue of speech waiting to be reprocessed |
| Playing back the moment a decision was made | Nothing was kept to play | A quoted line in the note, marked as a quote, with the caveat that it came from a transcript |
| Voice identification of a returning speaker | No voiceprint is created, deliberately | Speaker labels within a single meeting, discarded with the recording |
| Training a house model on real meetings | Nothing is retained to train on | Whatever the underlying transcription service is good at, described honestly rather than dressed up as ours |
Before any feature is added we ask one question. Does this require keeping audio after the note exists? If it does, the feature does not get built, however much easier it would make the product to sell.
The hardest part
Everybody else in the meeting
The person who presses record has consented to something. The other seven people have not, and no amount of product design changes that.
This is the problem that voice note taking tools mostly handle by ignoring. The user agrees to terms, the user turns on recording, and the other participants find out later or never. Their words end up in a transcript on a server belonging to a company they have never heard of, under a policy they were never shown.
We cannot solve that with software, and we are not going to claim we have. What we can do is refuse to make it worse.
- Nothing is hidden. There is no discreet mode, no minimised indicator and no way to record a meeting without the fact being obvious in the room.
- Nothing is retained about them. Once the note exists, the recording of their voice is destroyed along with yours. No voiceprint is built, so nobody in that meeting becomes findable in a future recording.
- The obligation stays with you. You are the one who has to tell the room, and in several Australian states you are the one who needs everybody to agree before you start. The privacy policy names the statute that applies in each state and territory.
- Third party requests are answered. If somebody who was in a meeting writes to us about their own information, we will deal with them as an individual with rights rather than as somebody else's data.
Status
What exists, stated without decoration
A company registered in 2026 with nothing released is not a failure. Describing it as anything else would be.
AMARA AI PTY LTD was registered in New South Wales in 2026. It has published no product, has no customers, has earned no revenue, has taken no outside investment and employs nobody it is willing to describe as a team on a web page. The work so far is a design, a set of retention commitments and internal experiments that have never processed anybody else's speech.
What would have to be true before anything is released
- Deletion has to be provable rather than promised. We want to be able to describe exactly what is removed, when, and what evidence exists that it happened.
- The consent flow has to be good enough that telling the room is the path of least resistance rather than an extra step people learn to skip.
- The note has to be better than the transcript by a margin that a sceptical person notices in the first meeting.
- The processing location and the sub-processor have to be nameable in public, because a privacy policy that says "our trusted partners" is not a privacy policy.
AMARA AI PTY LTD does not hold ISO/IEC 27001 certification, a SOC 2 Type I or Type II report, an IRAP assessment, an Essential Eight maturity attestation, HIPAA alignment or any other independent security or privacy accreditation, and will not represent otherwise until one is genuinely held. Nobody has audited us. No third party has tested our code.
The verifiable facts
Everything below can be checked against an Australian public register without asking us.
- Legal name
- AMARA AI PTY LTD
- Entity type
- Australian proprietary company
- ACN
- 696 682 827
- ABN
- 73 696 682 827
- State
- New South Wales
- GST
- Registered
- Registered
- 2026. Nothing has been published since
- Trading name
- Amara AI is a trading name of AMARA AI PTY LTD
- Verification
- The ABN, its current status and the GST position are published free of charge at abr.business.gov.au. The ACN sits on the register maintained by the Australian Securities and Investments Commission
- Service of documents
- The registered office recorded against ACN 696 682 827 at ASIC is the address that has legal effect for service. We do not print a second address here that would not have that effect
Name collision
We are not AmaraAI at amaraai.com
Two companies, similar names, no connection. This section exists so that nobody has to guess.
There is an established and unrelated company trading as AmaraAI at amaraai.com. They were using that name before this company existed. We are not affiliated with them, we are not a subsidiary, a partner, a reseller, a licensee or a successor, and we share no owners, no directors, no staff and no investors with them.
Anything you may have read about AmaraAI, good or bad, is about a different company unless it names AMARA AI PTY LTD or ACN 696 682 827. If you are looking for their product, their support, their pricing or their people, this site cannot help you and amaraai.com can.
We have not copied their positioning, their palette or their marketing, and we would rather lose the search traffic than collect a visitor who thinks they are somewhere else. If our similarity to them causes you a real problem, write to [email protected] and say so plainly.