Tell Mara the stakes. During the conversation she watches the pattern. Afterward she tells you what really happened, with the exact moments and the next move. The more she knows going in, the sharper the read.
Step oneWhat kind of conversation
Select all that apply. It lets Mara measure you against what this moment actually asks of you.
Step twoWho, and what is at stake
A few lines here change the whole read. All optional.
Who is it with
Their role or seniority
Their relationship to you
How important is this conversation
Step threeThe brief
Type, or press the mic on any box and speak. Mara holds all of this for this session only.
What is the meeting about
What outcome do you want
What are you worried about
Anything about this person or dynamic Mara should know
What do you want Mara to watch for
Step fourThe pressure, and how you want it told
Is this high pressure for you, and why
What do you need from this meeting
Pick the one that matters most. Mara measures the conversation against this, not against a generic ideal.
What are you most likely to do when this gets difficult
Your own read on yourself. Pick up to two. It tells Mara exactly what to watch for, and afterwards she tells you whether it actually happened.
What should Mara give you feedback on
How should Mara speak to you in the report
Step fiveHow Mara observes
Over a video call
Both voices. You speak on your microphone, the other person comes from the call's own audio. On start you choose your meeting window and tick Share audio. Best speaker separation. Needs Chrome 141 or newer.
In the room, your mic
For a conversation in person. Your microphone is read on your device. After a quick voice calibration, Mara separates you from the other person as best she can and tells you her confidence.
Try a sample
Runs a short scripted exchange so you can see a full report without a real conversation.
Read my voice and deliveryPace, pitch, energy and pauses from your microphone, on your device.
Read my presence on camera ExperimentalUses your camera on your device to read stillness, settling and fidget as a rough presence signal. Early and low confidence. Mara will say so. The video never leaves your device and is never stored.
How much should Mara say while you are in the room
She is silent by default and only speaks when a pattern holds. She never stacks prompts on top of each other.
Step sixTrain Mara to your voice
This is what lets Mara tell you apart from everyone else in the room. Train your voice once, tell her who else is there, and she labels each speaker by the sound of their voice instead of guessing. Thirty seconds, well worth it.
How many people will speak, including you
Who else is there, optional
Mara will not put a name on a voice on her own. During the conversation the others are Voice one and Voice two. When it ends she shows you what each voice said and you tell her who was who, so every name in your report is one you confirmed.
Voice not trained yet.
Microphone
Say a few words.
Camera
Camera off.
Press start, then say this line in your natural speaking voice. Mara learns your voiceprint on your device so she can pick you out from the other speakers.
Hi, I am here, this is my voice, clear and steady, and this is how I sound when I speak.
Voice not trained yet.
What happens to the conversation
Mara analyses while you are in session. Your words and theirs are turned into text on the spot, read, and dropped. By default nothing is recorded and no conversation is stored. When the session ends, Mara writes your report and the conversation is gone.
The report is the only thing kept, and it is yours. You can delete any report at any time.
Voice and camera are processed on your device. Audio and video are never stored unless you explicitly turn on recording.
Your individual meeting content is never shown to an employer or HR. Cohort dashboards show aggregate patterns only, never your words.
Before you startEveryone in the room
The people in this conversation need to know Mara is listening. Mara keeps no recording. Your words are turned into text as you speak by your browser's own speech recognition, which in Chrome is operated by Google. Over a video call, short pieces of the call's audio also go to our transcription provider, OpenAI, so that Mara can hear the other person. The text is held in this browser while you talk and is dropped once your private report is written. Nobody else sees your report, including your employer.
I have told the people in this conversationRequired before Mara will listen to a real conversation. The sample does not need it.
Mara will ask for your microphone, and for a video call, your meeting window.
00:00Observing. Nothing is recorded.
You · Them
··
Pace
··
State
··
Goal
··
Presence
··
Observing. What is said will appear here while it happens.
Who was who.
Mara separated the voices she heard but she will not put a name to one on her own. Tell her who was who and every name in your report is one you confirmed. Skip any you are not sure about and she will simply call them a voice in the room.
One last thing.
Before I write this up, was there a single moment you want me to look at most closely? Optional.
How did that go?
Mara is reading the conversation back.
Your report.
Let us talk about it.
Mara was in the room. Before she tells you what she noticed, she wants to hear how it felt.
Your communication pattern.
Loading...
Your best moments.
The lines where you were at your best, kept so you can reread your own proof of range before a hard meeting.
What should Mara watch.
Optional standing goals. Mara keeps these in mind in every report, on top of what you tell her before each meeting. For example, watch whether I close clearly, or watch whether I stop over explaining.
Your reports.
Privacy and data.
You own your reports. Here is exactly what Mara keeps, and your control over it.
How the Observer handles your data
Conversations are read in memory and discarded. Transcripts are never stored.
Audio and video are processed on your device and are never stored, unless you explicitly enable recording.
Reports are private to you. Your individual meeting content is never exposed to an employer or HR.
Where the Observer is used inside an organisation, cohort dashboards show aggregate patterns only, never any individual's words or reports.
If audio or video cannot be processed safely on your device, Mara falls back to report only mode and tells you.
Delete reports
Remove any single report from Past reports, or clear all of them here. This cannot be undone.
Your voiceprint
Mara keeps a numeric voiceprint, never any audio, so you train your voice once and she remembers you. Forget it any time and you will be asked to train again.