River Ranx

Building Folio — a companion for your AI Agents.

Agents create. Folio turns their output into something you can read, review, organize, and keep.

↓↓↓↓↓↓↓↓

[River Ranx @RiverRanx]

Agents like Claude Code and Codex generate huge amounts of Markdown every day. Yet I couldn’t find a Markdown editor that was both beautiful and effortless to use—something that lets you read and edit Markdown instantly, without having to learn the syntax.

So I built one.

[River Ranx @RiverRanx]

everyone on X is either scared of AI or hyped about it. 17.8% of working-age people on earth have actually used it. even in the US, where about half of adults use AI chatbots, only 17-20% of businesses use AI in any business function. i think the opportunity is in the other 82%

[River Ranx @RiverRanx]

@AnkanXplorer I have always used CLI because it can be integrated in the way you want, in Linux, in your own products, and has the same functionality as Desktop.

[River Ranx @RiverRanx]

@riffydays Thanks, that means a lot 🙏 I’ve spent a lot of time making the editor look good and feel great to use, and I’ve also added support for common Microsoft Office formats. Really happy to hear you appreciate it!

What part do you find yourself tweaking the most?

[River Ranx @RiverRanx]

Agents like Claude Code and Codex generate huge amounts of Markdown every day. Yet I couldn’t find a Markdown editor that was both beautiful and effortless to use—something that lets you read and edit Markdown instantly, without having to learn the syntax.

So I built one.

[River Ranx @RiverRanx]

@thedatadr1ver glad it caught your eye! it's built around docs in folders (markdown, code, docx, pdf and more) and does git diffs too. the full version, out in a few days, adds agent support like claude, codex CLI and antigravity

[River Ranx @RiverRanx]

@oliveiraacc "agents never dispatch benchmarks directly" is the kind of rule you only write after two of them fight over the same gpu at 3am lol. a tiny app whose whole job is making agents wait their turn is underrated

[River Ranx @RiverRanx]

@oliveiraacc great eval, and running it on both an m5 max and a 3090 is what makes it useful. ~30s of prefill every turn is where the whole loop quietly dies

[River Ranx @RiverRanx]

@oliveiraacc @davidfowl vscode is where i read markdown too, and it always looks like a code editor wearing a document costume. so i built folio, a mac app that opens any .md right away, mermaid renders, no vault, and it's actually nice to look at. free on the mac app store

[River Ranx @RiverRanx]

@riffydays the dyslexia joke is too real 😅 i stopped reading agent messages in the chat. anything long goes into a md file and i read it rendered, the wall of text is half the problem

[River Ranx @RiverRanx]

@Leex279 ended up with both. md for stuff i'll edit by hand, html when i just need to read it. 21x tokens for a drag and drop board is wild though lol

[River Ranx @RiverRanx]

@gregoriombp @shadcn "looks off" is the prompt version of telling a contractor the wall feels weird lol. pinning what changed is a much better habit

[River Ranx @RiverRanx]

@byte_array "a markdown editor reduces friction" is such a 2026 sentence lol. agents write the docs now and we're all just fixing one line at a time

[River Ranx @RiverRanx]

@CalvindeFox @shanebarakat Most honest answer in this whole thread lol. Building something and then asking your agent to give you the tour is a legit workflow.

[River Ranx @RiverRanx]

@gregoriombp looks really awesome

[River Ranx @RiverRanx]

@jaseempaloth every single line.

code handles payments in prod, so one hallucinated bug and we’re screwed lol.

codex and claude are honestly terrible for review, got fed up and built a mac editor just to read code comfortably. tossed it on the app store for free

[River Ranx @RiverRanx]

@RolloLucas "Totally. In B2B, slapping 'AI-powered' on your pitch now just invites procurement headaches and security reviews. Just show them what actually gets solved."

[River Ranx @RiverRanx]

telling devs to "just read the wall of text" fails every afternoon.

the real fix isn't willpower. it's layered review:

1. zoom out: scan decision diffs first
2. zoom in: read code only where intent feels risky
3. fix plan flaws in-place

stop reviewing outputs in a terminal

[shadcn @shadcn]

The most important coding skill: read.

Teach yourself how to read.

Read the agent output. Read the blog post. Read the paper. Read the code (if it applies to you).

Don't summarize.