Meera
Raises concerns as questions, and lets them drop when they aren't picked up.
Twice she offered a way into the QA problem, and both times the conversation moved on. A direct question back may get you the whole concern.
Octo · a conversation lab · crowdfunding after the beta
Hear how you show up.
Octo records your conversations, works out who said what, and reviews each one against principles you write yourself. Then it suggests one or two small things to do differently next time.
It runs on your own Mac. The audio never leaves it.
Example
What it does 10 m
Three steps, then your part.
-
1
Listens
A small recorder on the table, or the voice memos on your phone. The recorder shows when it's recording, on its screen and with a soft pulse of light.
-
2
Understands
On your Mac, Octo finds each conversation, writes down every word and works out who said what. It recognises voices you've named before.
-
3
Reflects
A review of the whole conversation: how the others seemed, what went unsaid, how you did against your principles, and one or two small next steps with a date on them.
Then your part
You confirm who's who, mark what was new to you, and tick what you acted on. Over weeks, a coach reads across your reviews and names the one habit most worth working on.
- Confirm who's who
- Is it right?
- Is it new to you?
- Did you act on it?
How it works 30 m
Eight arms, eight jobs.
About two-thirds of an octopus's neurons are in its arms, and each arm can act on its own. Octo works the same way. Each of its eight steps runs on its own and does one job. You are the head. The review waits until it's clear who's who, and you decide what it got right.
Follow one made-up conversation through all eight. It's a 1:1 between you and Meera, a team lead, about a launch date.
Arm 1 of 8
Record
The recorder sits in the room and records onto its memory card in one-minute pieces. When it's on your home Wi-Fi, your Mac collects the pieces every five minutes. Voice memos from your phone can go in too.
Under the hood
A Seeed XIAO ESP32-S3 Sense board with its own microphone, recording at 16 kHz. Each piece is encrypted on the card so only your Mac can read it. Every request between the Mac and the recorder is signed, and a piece is deleted from the card only after its SHA-256 checksum matches on the Mac. Recordings are kept as FLAC, which is lossless and about a third the size of WAV.
LimitAcross a room, voices arrive 15 to 35 dB quieter than on a phone held close. People still hear every word; the models start missing them sooner.
Arm 2 of 8
Find conversations
A day of recording is mostly silence. Octo skips it and keeps the stretches where people talk. Three minutes of quiet ends a conversation.
Under the hood
Silero voice detection runs on a louder copy, after an 80 Hz high-pass filter. The board's microphone drifts below 5 Hz, and until that was filtered out the detector found no speech at all. A conversation needs at least 2 seconds of speech.
LimitTwo meetings back to back can come out as one. You can split them where the people change.
Arm 3 of 8
Write it down
Every word, with the moment it was said. First Octo works out the language, then it writes the words.
Under the hood
Whisper large-v3 votes on the language over the five most speech-heavy stretches, choosing among English, Hindi, Kannada and Gujarati. Qwen3-ASR-1.7B writes English; it beat Whisper 7 to 3 in a blind test on real conversations. For Hindi and mixed conversations, both models write a draft and the review's model merges the two from their text. That won 6 of 9 in a blind test, with no invented words.
LimitKannada and Gujarati are paused. Names of people and places are the weakest spot for every model tried.
Arm 4 of 8
Who spoke when
Octo splits the sound into voices, A and B here, and marks when each one is talking. It doesn't know their names yet.
Under the hood
pyannote 3.1 on the Mac's GPU, with a clustering threshold of 0.705. Two newer models were tried on real recordings and each lost a person, putting two people on one voice, so 3.1 stayed.
LimitSimilar voices can blend. One voice split in two takes one click to merge; two people on one voice take longer to fix.
Arm 5 of 8
Who said what
Each word goes to the voice that was talking at that moment, and the words become lines in a transcript.
Under the hood
Words are placed by how much they overlap each speaker's turns, and short gaps are filled by comparing voices. A line whose speaker is uncertain gets a ⚠, so nobody builds a conclusion on it unknowingly.
LimitIn fast back-and-forth, a word can land on the previous speaker's line.
Arm 6 of 8
Whose voice
Octo compares each voice with the people you've named before and suggests a name for each. It accepts your own voice on its own; other names wait for your yes.
Under the hood
Voiceprints come from WeSpeaker ResNet293. A name is suggested at a similarity of 0.60 or more, at least 0.15 ahead of the next person. The owner's own voice is accepted on its own at 0.70 with a 0.25 lead, and adds no voiceprint until they confirm it. Prints from under 8 seconds of speech are ignored, and each person keeps at most 12.
LimitSomeone else gets a voiceprint only once you record that they agreed. Without it, you type their name by hand.
Speaker AYouauto0.86 · 0.41
Accepted on its own. Confirming adds a voiceprint.
Speaker BMeera?0.71 · 0.38
Once A is you, A turns blue: you are always the accent colour in the app. Try the buttons.
Arm 7 of 8
Count
Talk share, turns, questions, interruptions and filler words, for everyone in the conversation.
Under the hood
Plain arithmetic, no AI. Questions come from punctuation and interruptions from overlapping speech, so both are marked approximate. Filler words are counted only where the transcription kept them.
LimitNumbers are never coloured good or bad. The review uses them where they say something.
Arm 8 of 8
Review
A model you choose reads the whole conversation and writes a review: how the others seemed, what went unsaid, how you did against each principle, and what to do next.
Under the hood
Claude Opus 5.5 on your own Anthropic key, any OpenAI-compatible service on yours, or a local model through Ollama. Only the transcript text, names and numbers are sent, and only to the service you chose. Every quote is checked in code against the line it cites, and an item that loses its evidence is dropped. A reading of someone else needs two of their own lines from different moments.
LimitTone is inferred from words, pauses and turn-taking, and the review says so when it does.
Between the lines
Her 03:12 question never got an answer. The date was held at 17:40 without coming back to QA.
What to do · next 1:1
Read the whole review
Ask Meera what QA needs to hold the date, and hear the whole answer first.
1 of 8
Then the head. You answer any suggested names (the review waits for that), mark what was new and what you did. Once you've answered a review it stays as it is, even if the transcript changes later, unless you ask for a redo.
What you get 60 m
Feedback the way a perceptive colleague would give it.
An example. The conversation, the names and every line are made up to show the shape of a real review, read against a set of example principles.
The brief every review follows: review the conversation I had with others and give me feedback the way a perceptive colleague would
127.0.0.1:8765 · Conversation labExample
1:1 with Meera
6 to rateKeys work on the highlighted card:R P W right, partly, wrongN M new, knew it
Summary
A 24-minute 1:1 about the launch date. Meera raised QA capacity twice; the date was held without that being answered. You took her frustration about hearing things last well. You did most of the talking (68%).
Reading Meera
To rate
Between the lines
Moments 14 minutes apart
Her 03:12 question never got an answer.
You held the date at 17:40 without coming back to QA's two missing people. Her “Sure. Makes sense.” came after a four-second pause, so it may be agreement on the surface only. That reading is inferred from the pause and the short reply.
Your conduct
Principle 11
Be friendly with everyone, but choose who you get close to.
When Meera said the team always finds out last, you answered “That's fair. Tell me more.” and let her finish. That's valuing the colleague who gives you honest feedback.
Principle 12
Watch your own four faults: anger, jealousy, deceit and ego.
At 21:10 you leaned on your track record, and the QA question closed there. Whether ego was part of it only you can tell; the transcript can't show what you were thinking. The effect was that her concern went unanswered.
No clear moment for principles 1 to 10 or 13 in this conversation, so nothing is said about them.
What to do
Ask Meera what QA needs to hold the date
Next 1:1Open your next 1:1 with her question from 03:12, and hear the whole answer before you suggest anything.
Why: she raised it twice and it was never answered.
Example only: nothing was sent. The real button sends the title, details, due date and a link back.
Take her up on the QA write-up
FriShe offered one at 11:05. Ask for it this week and read it before the steering meeting.
Why: it turns her concern into something you can act on.
Example only: nothing was sent. The real button sends the title, details, due date and a link back.
Limits. Lines from 08:40 to 09:10 have an unsure speaker (⚠). Tone is inferred from the words, pauses and turn-taking. Nothing here can say what either of you was thinking.
Every quote is checked in code against the line it cites. An item that loses its evidence is dropped before you see it.
Why Octo 90 m
An octopus has three hearts. Octo has three goals.
Octo began in September 2026 as a tool for reviewing one's own conversations. These were its goals, and they still are.
-
Get better
Feedback on how you actually show up in conversations.
-
Learn from others
From people who have already solved the problems you are facing now.
Not built yet -
Grow, by your own standard
In life and career, with conduct measured against your own principles rather than generic management advice.
The standard
Your principles first. Career advice only if it passes them.
Most coaching measures you against getting ahead: visibility, positioning, influence. Your own standard may ask for something else. When the two disagree about the same conversation, Octo goes with yours.
Every conversation is judged against your principles first. Career feedback still comes, but only when it fits them, and you can name advice you never want given: to claim the credit, position yourself or win the room, say. That advice is then never given, not even softened.
The filter, on the example 1:1
Advice a typical coach might give after that conversation, checked against that rule:
keep a suggestion only if it passes every conduct principle. If a useful career point would conflict, leave it out rather than soften it.
- Send Meera a short note of what was decided and what's still open.Kept.
- Announce the held date in the steering meeting yourself, so leadership sees you making the call.Left out: “Never positioning myself, winning the room or taking credit.”
- Ask Meera what QA needs before the date is final.Kept.
- Open the next meeting with your three past launches, to set your authority.Left out, for the same reason. Not softened into a gentler version.
The principles
A few lines in your own words. Every conversation is held to them.
Three, from the principles Octo was first built around:
- 3
Integrity is what makes people trust you.
- 7
Stay calm; hostility is not the answer.
- 12
Watch your own four faults: anger, jealousy, deceit and ego.
When you set up Octo, you write yours, from whatever you measure yourself by. The review reads them exactly as written. Octo never writes them, edits them or suggests any, and neither do we.
Your voice 120 m
Your Mac, and the people you talk to.
Octo is for one person, but most conversations have at least two. This is exactly what happens to everyone's voice and words. There are no accounts and no servers of ours: nothing you record, write or set ever reaches us.
Records
Recorder or phone
Shows when it's recording. Sends pieces only over home Wi-Fi, signed, checked before they leave the card.
Audio never leaves this box
The owner's Mac
Kept as FLAC, and deleted 60 days after it's added.
Transcript text · names · numbers
The review model you chose
Only for conversations you're in, once the speakers are settled. On your own key, or a local model so nothing leaves at all.
A task, if you connect it and tap Add
TickTick
Title, details, due date and a link back.
- 01
You'll know when it's recording.
The recorder's screen shows the octopus listening, with REC and the time, and its light pulses every five seconds while it records. There's no hidden mode.
- 02
Voices stay on the Mac.
Audio, voiceprints, transcription and telling voices apart all happen on the owner's Mac. No cloud service hears anything, and that isn't a setting.
- 03
Only text leaves, and only to the review model.
The transcript text, names and numbers, on the owner's own key; or nothing at all with a local model. Only for conversations the owner is in. One they aren't in gets no review, just numbers.
- 04
Voiceprints only with consent.
Someone else's voiceprint is kept only once the owner records that they agreed: when, and how. Without it, names are typed by hand and no voiceprint is made.
- 05
Audio goes after 60 days.
Transcripts and reviews stay, on the Mac. Deletes reach its backups within 14 days.
- 06
The review is about the owner.
It describes what others did and seemed to want, in their own words, so the owner can understand them and respond well. Never a verdict on anyone's character or worth.
What anyone in a conversation can ask for
Anyone recorded can ask the owner for any of these, at any time. Each is a button in the app, and each asks before it acts.
Choose one to see what it removes.
The recorder 150 m
A small box with an octopus on its screen.
A microphone, a memory card, home Wi-Fi and a small screen, in a two-level case that sits on the table in plain view. It records from the moment it's switched on, until you stop it. This one works: try the controls on its side.
Starting
Tap record to wake the screen; hold it for 2 seconds to start or stop. Flip the power switch to turn it off and on. After a few seconds the screen dims, the way the real one does.
A drawing of the case design, not a photo. M1 is the Seeed board and EB the expansion board, which carries the screen. M1 stands on EB, which makes one end taller. The lower level's top is dark glass, with the screen and the recording light under it. The working recorder runs today as bare boards on a power bank; the case is designed and not yet printed.
- Board
- Seeed XIAO ESP32-S3 Sense (M1), 8 MB of PSRAM, standing on Seeed's expansion board (EB) with its screen and battery socket.
- Microphone
- On M1, 16 kHz.
- Card
- 32 GB microSD, written in 60-second pieces, each encrypted so only the Mac can read it.
- Home
- The Mac collects pieces every 5 minutes over home Wi-Fi. Every request is signed, and a piece is deleted from the card only after its SHA-256 matches.
- Screen
- 128 × 64 OLED, under dark glass on top of the case. Bright for 6 seconds after anything happens, then dim, redrawn once a second and shifted a few pixels each minute so nothing burns in.
- Light
- Under the glass, just behind the screen. A soft pulse every 5 seconds while recording, and a double pulse every 3 seconds when it needs attention.
- Side
- Record: tap to show the state, hold 2 seconds to start or stop. Power: a switch, so off is really off.
- Phone
- An iPhone app finds it over Bluetooth, pairs with a six-digit code from its screen, adds Wi-Fi networks and starts or stops recording on the go. Android later.
- Updates
- Over Wi-Fi or USB, signed, and refused while it's recording. If new firmware doesn't start, it goes back to the old one on its own and says so.
The screen on this page uses the recorder's own drawing code, ported line by line and checked against the original: 1,080 frames, not one pixel different. On a 128 × 64 screen the octopus makes do with six arms.
The plan 170 m
One person first. Then ten. Then a campaign.
Octo goes to crowdfunding only once it has shown it works: first for us, then for people who didn't build it. Each stage has a bar written down before the results come in.
-
Now, since 1 Oct
Our own test
Does feedback on our own conversations tell us anything we didn't already know? It passes with 10 conversations reviewed, at least 1 in 3 with something new, and 2 things acted on.
-
Next, after it passes
A beta with 5 to 10 people
On their own Apple Silicon Macs, with their own principles and their own model key, using phone voice memos for four weeks. It passes if they install it without help, at least 1 in 3 of their reviews has something new to them, and they're still using it in week 4.
-
Then, after the beta
The crowdfunding campaign
For the recorder, assembled in its case. Before it opens: a printed case that fits, certification for the boards, and a production plan for a first batch, shown on a working prototype.
Where stage one stands
If it passes
The beta begins, and the case gets printed and fitted.
If it fails
We stop building, and there is no campaign.
What backers would get
- The recorderAssembled and tested, in its two-level table case, ready to pair.
- The lab for your MacThe whole pipeline, from audio to review, with a one-step install.
- The phone appFor the recorder on the go, and your lab's pages from home or anywhere.
- All of it open sourceThe software, the firmware and the case. The software is free; the recorder is what's sold.
What you'll need
- An Apple Silicon Mac with 16 GBTranscription and voices run on its own chip. No Windows or Linux in the first version.
- A model for the reviewYour own Anthropic key, any OpenAI-compatible service, or a local model.
- An iPhone, for the appOptional. The recorder also pairs over USB from the Mac.
- Your principlesA few lines in your own words. The review is only as good as the standard you give it.
What it can't do yet
- Kannada and GujaratiPaused. Hindi works through two drafts merged into one; English is the strongest.
- Voices across a roomQuiet audio is where models fail first. People hear through it; models don't.
- Similar voicesThey can blend into one speaker. Names stay suggestions for that reason.
- Tone of voiceRead from words, pauses and turn-taking, never from the sound itself.
Not decided yet
- The platformWhich crowdfunding site, and who it ships to.
- The priceSet once the case, the parts and certification are costed.
- The licenceChosen before the code goes public, so a closed copy can't drop the privacy promise.
- AndroidThe app is built to run on it; iPhone comes first.
Follow along 190 m
Hear when the beta opens, and when the campaign does.
Two kinds of news, rarely: a place in the beta if you have an Apple Silicon Mac, and the day the campaign opens.
Tell us you're interested
This copy of the page can't keep a list. Write to [email protected] and we'll add you.
Sign-ups so far0
Only the page's owner sees this.