A complete guide to voice journaling in public using AI

You can keep a journal by talking to your computer. You say what happened, an agent/transcriber tidies it into something readable, and it lands on your blog as a draft. Later you read it, fix what you want, and decide whether it stays a draft or goes up.

This post is the whole setup. What voice journaling is, why a blog works as a journal, and how to wire it together so the words end up on lykhari.

What voice journaling is

Journaling by speaking instead of typing. That is the whole idea.

It works because talking is easier than writing. A blank page asks you to be articulate before you have worked out what you think. Talking asks for nothing. You can start in the middle, contradict yourself, trail off, and come back.

Then something has to turn the talking into text. Your computer already does this. On Windows it is Windows + H. On a Mac, turn on Dictation in Keyboard settings and use the shortcut it gives you. Both put words wherever your cursor is.

The part that used to be missing is what happens after. Raw dictation is a mess. You get a wall of half sentences with no punctuation, and reading it back is worse than having written nothing. An agent fixes that. It takes the mess and hands back a paragraph that still sounds like you.

What lykhari is

lykhari is a blogging platform I built. You write, it publishes, you own the words. The free tier gives you a blog at yourname.lykhari.com. The paid one is $5 a month or $50 a year, and adds custom domains, a newsletter, comments, themes, and the API this setup depends on.

It is deliberately small. No algorithm, no feed deciding who sees your posts, no recommendation network pushing you to post more. Your blog is your blog.

Keeping a journal in public

Something worth saying plainly before you start. A lykhari blog is on the web. Anyone with the address can read it. Do not put anything in there you would be upset about a stranger reading.

That middle ground turns out to be useful, and there are a few ways to use it. A work log, what you built and what broke, with the link sent to two colleagues. A reading journal. Morning pages you never link to anywhere, treating the blog as a drawer that happens to sync. Or writing in the open on purpose, because knowing one person might read it is what makes you finish the sentence.

If you want a genuinely private diary, use a file on your computer. This is for the writing you would not mind being read.

Setting it up

Four steps. The middle two take about five minutes.

Create a blog. Go to lykhari.com, put in your email, click the link it sends you, pick a name. That is the blog done.

Switch to pro. The API is a paid feature. Dashboard, billing, pick monthly or yearly. You get other options too. Like customization, emailing to your readers, etc.

Generate a token. Dashboard, settings, search for "api", generate a personal access token. You see the full value once, right after you create it, so copy it somewhere safe. You get one active token at a time and you can revoke it whenever you want.

Hand the token to your agent. This is the part people overthink. You paste three lines:

blog url: https://yourname.lykhari.com
token: lykh_your_token_here
api docs: https://blog.lykhari.com/api-docs

That is enough for Claude Code, or for anything else that can read a page and make an HTTP request. It reads the docs, sees the shape of a request, and works out the rest. You do not need to explain the API to it.

A prompt to start with

blog url: https://yourname.lykhari.com
token: lykh_your_token_here
api docs: https://blog.lykhari.com/api-docs

I am going to talk for a few minutes. Turn it into a journal entry in my
voice. Keep my phrasing where it is fine. Do not add anything I did not
say. Send it to my blog as a draft, titled with today's date.

Then press the mic button in claude code and talk.

The mistake worth avoiding is dictating a finished entry. Dictate the raw material instead. What happened, what you thought about it, what you would say to a friend over coffee. The writing gets easier once the thinking is out of your head.

What happens to the audio

Nothing reaches lykhari. There is no audio field in the API, no upload, nothing stored. Your speech becomes text before any of this touches the blog, and only that text gets sent.

Whatever you use to turn speech into text has its own policy, and that is worth reading. Claude code dictation, Windows dictation, Mac dictation each handle it differently. But the blog end of this is text and only text.

What else the token can do

Publishing is the obvious use. The token does considerably more, and once it is wired up you can ask for these out loud.

  • Drafts. Send everything as a draft. Nothing goes live until you have read it.
  • Reading your own writing back. Drafts included. "What did I write about last Tuesday" works, and so does "find the entry where I talked about the deploy going wrong".
  • Editing. Retitle something, rewrite a paragraph, change the slug or the description.
  • Publishing and unpublishing. Push a draft live, or take down something you have thought better of.
  • Scheduling. Give it a date in the future and it goes out on its own.
  • Pages and category pages, not only posts. Your about page can come from the same place.
  • Pinning a post to your nav.
  • Emailing your subscribers when you publish, if you have a newsletter going.

What it cannot do, deliberately:

  • Delete. The token reads and writes. Deleting a post needs you, signed in to the dashboard.
  • Touch anything else. Settings, comments, members, billing and your account are all dashboard only. A leaked token cannot cost you money or expose your readers.
  • More than ten new posts a day. There is a cap on posts created with a token.

Ten is more than most people write. If you are producing more than ten journal entries in a day, the cap is probably not the thing to fix.

Where this falls down

Agents flatten your voice if you let them. They round off the odd phrasing that made it yours. Saying "keep my phrasing" in the prompt helps, and reading the draft before publishing helps more.

And this only makes sense if you already have the agent open. If you do not use Claude Code or anything like it, setting one up purely to journal is a lot of machinery for a paragraph a day. Open the lykhari editor and type. That works fine.

Why it is worth setting up

The entries that never get written are rarely blocked on typing speed. They are blocked on sitting down. Talking removes the sitting down. By the time there is a draft to look at, the hard part is done, and what is left is deciding whether you meant it.