Azynote 1.3.0: Free, Local, and Not Just for Meetings
Three things changed in this release. The app is no longer only about meetings, the AI now runs on your Mac, and Azynote is free.
They look like three announcements. They are really one decision, seen from three sides.
From meeting companion to work companion
I built Azynote because meetings were where I lost the most information. That framing was right for a while, and it quietly became a cage. I kept adding things I needed during the rest of the day, then describing the app as if those things did not exist.
So here is the honest version of what a day with Azynote looks like now.
9:12, writing a reply you do not want to get wrong. You select the text in whatever app you are in, Mail, Slack, a Jira comment, and press Ctrl+Shift+O. Proofread it, tighten it, translate it, or run one of your own actions that you defined once and now use every day. The transformed text comes back where you were. Azynote never asked you to come to it.
11:40, a thought lands while you are in the middle of something else. Ctrl+Shift+K drops it into your workspace and gives you back what you were doing. No new window to hunt for, no note file to name, no context switch you have to pay for twice.
14:00, the meeting. This part has not changed and it is still the anchor of the product. Transcription runs on your Mac, speaker attribution runs on your Mac, and when it ends you get a OnePager with the summary, the decisions, and the tasks that came out of it.
16:30, the question you cannot answer from memory. Ctrl+Shift+C opens chat over your entire workspace, not only your meeting history. What did we decide about the pricing model. What is still open on this account. Which sessions touched this subject last quarter. And if you connect an MCP server or two, the same conversation reaches your other tools without you leaving it.
Meetings are still the heaviest thing Azynote carries. But the app was already living in the rest of the day, and calling it a meeting companion was making that part invisible. It is a work companion. That is what the product page says now, and this release is the version where I stopped hedging about it.
A first real attempt at running everything locally
This is the part I have been building toward since the beginning, and I want to be precise about what has and has not arrived.
Transcription has been on-device since day one. Storage has always been plain files on your disk. But the intelligence, the step that turns a transcript into something you actually use, was a cloud call to Google. The most interesting part of the pipeline was the part that left your machine.
In 1.3.0 it does not have to. Azynote ships with a local model, downloads it for you, and manages it. It writes your OnePagers, powers chat and the Command Panel, and runs the quick text transforms. A second, smaller model handles search, so indexing and semantic search across your OnePagers work with no cloud provider configured at all. On a Mac with 16 GB of memory or more, you can run the entire product with nothing about your work leaving it.
The cost is honest and worth stating: about 6.8 GB of models to download once, 16 GB of memory as the floor, and output that is close to the cloud without being equal to it.
You are not locked into that choice. The first time you open a workspace, Azynote asks which engine should do the thinking:
- Local, the bundled on-device model, pre-selected unless your Mac has under 16 GB.
- Gemini, the cloud path, still the quality ceiling today.
- Any OpenAI-compatible provider, which covers cloud OpenAI and anything speaking that API, including a server you run yourself.
- The Claude CLI, if you already work with an agent in your terminal.
The choice is not all-or-nothing. In Settings > AI Configuration each capability routes on its own, so you can have OnePagers written locally, chat on Gemini, and search on-device, if that is the trade you want. Two capabilities have no local answer yet and say so plainly instead of pretending: voice and image generation are off when you pick Local.
Three things it does on its own, which you never have to configure. The context window is sized from the memory your machine actually has, so a 16 GB Mac gets a solid working window and a 64 GB Mac gets a very large one out of the same model. The model starts warming the moment you open chat, because a cold start is the slowest thing about running locally. And if a model file ever goes missing, Azynote notices at startup and offers to fetch it again instead of pretending it is there.
I called this a first attempt deliberately. Local models became good enough to be the default faster than I expected, and "good enough to be the default" is not the same as finished. Expect the local path to be slower on the first prompt of a session and occasionally less polished on a hard summary. It is real, it works, and it is where the product is going.
Azynote is free, and I will refund you
No trial, no tiers, no account, no card. Every feature, for everyone.
This is not me admitting the paid version was not worth it. It worked, people did real work with it, and I would defend what they paid for. What changed is not what the app is worth. It is what I want this to be.
The honest framing is that Azynote is an exploration space. I am trying to find out what a work companion looks like when nothing has to leave your machine, and the hosting, the developer account and the hours are what that exploration costs me. I share the result with the people who recognise something of themselves in the idea. If that turns into a way to fund the work later, good. It is not the reason I am doing it.
Which puts this closer to a friend showing you something he built than a company shipping you a product, and it comes with the trade that implies. There is no roadmap I owe you and no support queue with a clock on it. Azynote moves at my rhythm, and I would rather say that plainly now than have you find out on the day you need something urgently. What you get in return is an app with nothing gated, no account, no analytics, and nothing of yours reaching me.
If you already bought a license, write to me at support@azynote.com and I will refund you. Your license still validates, and you now have exactly what everyone else has. The refund is not a formality I am hoping you skip. You paid for something I have since given away, and that gets made right.
The longer version of this reasoning is on the Pricing page.
Also in this release
A few things worth knowing if you are already using Azynote:
- OnePagers stopped reading your transcript twice, which was quietly doubling the work behind every summary.
- Output language holds. Summaries stop drifting back to English when you asked for something else.
- Tool failures say what failed instead of hiding behind "An internal error occurred".
- Inline
#!instructions now run through the OnePager engine rather than chat, so they respect the engine you configured.
Tell me where this is wrong
I have changed my mind in public about the price once already, so I am clearly reachable on this.
If the local output is not good enough for your work, I want to know on which kind of session. If the work companion framing sounds like a stretch from where you sit, tell me what you actually use the app for. If you bought a license, mail me and take the refund.
Happy meetings, and everything around them.