Bogdan Nichovski

Building tools I wish existed 🛠️
🧪 – A/B test any page, no code
✈️ – AI plans your trips in minutes

[Bogdan Nichovski @Nichovski]

Don't wait until you're 100% ready. Start now.

[Bogdan Nichovski @Nichovski]

Japan is next level. Just your casual, everyday Hello Kitty train rollin' up

[Bogdan Nichovski @Nichovski]

Travo, last 28 days:
→ 1,031 active users
→ 887 new
→ 5 paid · $66 MRR
Activation works. Monetization doesn't. Free → paid is the next mountain.

[Bogdan Nichovski @Nichovski]

@jocotoo @jacobrodri_ Interesting add-on for Travo😊

[Bogdan Nichovski @Nichovski]

Travo just hit 700 Google clicks/month from programmatic SEO.
Lesson: stop building 1 perfect landing page. Build 500 useful ones from a template and let Google sort them.

[Bogdan Nichovski @Nichovski]

@MichLieben Ads

[Bogdan Nichovski @Nichovski]

https://t.co/2Ob8dwNNQ0

[Bogdan Nichovski @Nichovski]

i was tired of watching 5-minute Looms at 2x just to find out what the bug was

so Claude does it now. paste the link, it transcribes locally with whisper, reads the transcript, goes straight to fixing. feels like cheating honestly

link in the comments:

[Bogdan Nichovski @Nichovski]

an agent overwrote one of my wiki pages with a placeholder yesterday.
no read_page in that session. just write_page. original content gone.
so i built page history into meshnote.
every write now auto-saves the previous version. agents can list_page_history and restore_page to undo the damage. no setup, no config. works on existing wikis from the first overwrite.
if you're letting LLMs write to your knowledge base, you need a safety net.
this is it.

[Bogdan Nichovski @Nichovski]

my experience with Opus 4.7 so far 🥴

[Bogdan Nichovski @Nichovski]

@PaulYacoubian The real one! 👏

[Bogdan Nichovski @Nichovski]

@MladenovskiJ ⏬️⏫️

[Bogdan Nichovski @Nichovski]

Claude is down, I repeat, Claude is down!

[Bogdan Nichovski @Nichovski]

@DeepakNesss Not for me, still down

[Bogdan Nichovski @Nichovski]

if you've ever opened claude code and thought "I wish this thing remembered what we figured out last week" you've already understood the core problem with LLM workflows.
the agent is brilliant for 20 minutes and amnesiac forever after.
spent the last few months building the fix. plain markdown files on disk, your agent maintains them across every conversation. no RAG, no vectors, no cloud.
npx meshnote init ./brain
https://t.co/4H3bO3pSeu

[Bogdan Nichovski @Nichovski]

@nonameblitz https://t.co/qsuHDxREEg

or if you want to skip the reading and just try it:

npx meshnote init ./brain --template research

[Bogdan Nichovski @Nichovski]

every time I watch a serious LLM user work, they end up building the same thing: a folder of markdown files and a claude md telling the AI how to maintain it.
karpathy did it. I did it twice. probably you too.
got tired of rebuilding it so I made it a product. meshnote.

[Bogdan Nichovski @Nichovski]

congrats on the year of work, that shows in the demo. and yeah, hatch and meshnote are clearly aiming at different shapes of this — hatch as a full workspace, meshnote as the thin layer for people who already live in their terminal and editor.
different audiences, same karpathy gravity well. good week to be in this category

[Bogdan Nichovski @Nichovski]

follow up to yesterday's post:
the second thing every serious LLM user builds is a cleanup script. because two weeks in, your wiki is full of duplicate facts and the agent starts contradicting itself.
someone in the replies called it the "cleanup loop." best name I've heard for it.
nobody talks about this part. but it's the thing that makes or breaks whether the second brain actually works.

[Bogdan Nichovski @Nichovski]

the whole loop:
npx meshnote init ./research
drop a paper in. tell claude "ingest this." come back to a summary, author pages, concept pages, all wikilinked into stuff you ingested last week.
no UI. no tags. you just read.