In the 1.3.0 post I wrote about Azynote outgrowing the meeting. It had quietly become a work companion: there at nine when you are writing a reply you do not want to get wrong, there at two for the meeting, there at half four for the question you cannot answer from memory.
Rereading that now, one piece is obviously missing.
Azynote knew everything you had written down and everything that had been said in front of it. Every note, every transcript, every OnePager, searchable and cross-referenced. That is a very good knowledge base about your work. It knew nothing whatsoever about you.
And a colleague who attended every meeting and remembers all of it is still not a colleague if you have to re-explain yourself to them every morning. That the platform team is the customer's and not yours. That a commitment you make in a meeting is a follow-up you owe rather than a task you are handing out. That you want the decision at the top and not three paragraphs of background above it.
The gap is between a knowledge base, which holds what happened, and a companion, which also knows who it is working for. 1.4.0 is the first step across it.
It learns who you are, not just what was said
Azynote now keeps a short profile of your world and applies it to everything it writes for you. It builds that profile from how you actually work, out of two places: the conversations you have in chat, and the OnePagers it generates.
Chat is where it learns you. Your role and which side of the table you sit on. What your words mean here, because "a POC" in your world is a paid engagement with written success criteria and in the next company it is a demo. How you want things written, because you keep moving the decision to the top and you keep deleting the background section.
Your OnePagers are where it learns your world. The people, the projects, the acronyms, the internal names nobody outside your company has ever typed. This is the part you feel first, because it is what stops a product name coming back for the fourth week running as three ordinary words that sound roughly like it.
It does not overreact to a single sentence. Asking for something shorter once is an instruction about that one document. Asking across three separate conversations is a preference, and only then does it get written down. That patience is deliberate: a companion that rewrote its entire approach every time you said something in passing would be exhausting.
It runs quietly when your machine is not busy, and never while you are transcribing. Nothing waits on it.
Or you can just tell it
Waiting for it to work you out is the slow way round. In chat, say so:
You: remember I always want action items at the top
Azynote: Got it.
That goes straight in and stays. No waiting, no repetition, and the background learning can never overwrite it or decide it has gone stale. This is the fastest way to be useful on day one, before it has watched you long enough to infer anything.
Ask "what do you remember about me?" to see the list. To change it, say that too. "Forget that I want action items at the top" removes it. "Actually Marie moved to Northwind" corrects the line she is on, rather than leaving two versions sitting there contradicting each other.
The same works from outside the app. If Azynote is connected to Claude Desktop, Cursor or anything else that speaks MCP, those can teach your workspace too. You are deep in something in another window, you learn who owns the schema registry, and it lands in Azynote without you switching apps to write it down.
You can read every word of it, and change any of it
This is the part I would want if somebody else had built it.
Open Settings then AI Configuration, and under your custom prompt there is a box called What the agent has learned. That is the entire profile, in plain English, and it is a text box. Delete a line you disagree with. Type in something it has not worked out yet.
That matters because it can be wrong about why. If you deleted three background sections because those three were bad, rather than because you dislike background sections, it may draw the wrong conclusion. Something that shapes everything Azynote writes for you should be a paragraph you can read in ten seconds and correct in twenty, not a model weight you have to argue with.
It is also a plain Markdown file in your workspace, at .azynote/workspace-profile.md. Open it in whatever you use. If your workspace is in git, you can see exactly what it learned and when.
Two buttons sit underneath. Learn now runs a pass immediately instead of waiting for a quiet moment. Forget everything wipes it and starts over.
It learns per workspace, so your consulting workspace and your side project each build their own picture of you. You are a different person in each, and one profile stretched across both would be wrong in both.
What that changes for you
The OnePager you stop rewriting. You have been making the same three edits after every meeting: pulling the decision up, cutting the background, splitting what you owe them from what they owe you. It starts arriving that way.
Names that survive the week. Your product, your internal tooling, the colleague whose surname the transcription has never once got right. Once Azynote has seen them written down properly, it keeps writing them properly.
Chat that knows the cast. Ask what is still open on an account and you get an answer that knows who those people are and which side they are on, rather than one that treats every name as a stranger.
It runs on whichever engine your workspace already uses, so if you are set to Local all of this happens on your Mac and nothing leaves it. If you are on Gemini, it goes where your chat already goes. And if you would rather it did not happen at all, set the Workspace profile row to Off in AI Configuration and the whole thing stops, box included.
Also in this release
- A window that fits beside your call. It shrinks to 320×400 and Ctrl+P pins it on top.
- Stealth mode you can work in. The navigation rail comes back from the left edge, and transcription controls appear when you hover the status dot.
- Your session's documents on tabs, with transcripts named by when they started.
- Tasks that say which meeting asked for them, on hover.
- Timeline one-liners you can actually read in a narrow window.
- Your custom prompt now reaches chat and voice. It was only reaching OnePagers.
I want to know what it gets wrong. Turn it on, give it a fortnight, then open that box and tell me which line made you laugh and which one you would have written yourself. And if you tell it to remember something and it lands somewhere odd, send me the sentence you used. I read everything at support@azynote.com.