re·grok /ˈrēˈgräk/ verb

Understand what the AI just wrote, at a glance.

to grok again: to understand, at a glance, the wall of text an AI just wrote.

An AI can write a spec, a plan, a diff, or five hundred lines of code in seconds, and it writes all of it as long text. regrok takes anything an AI writes and turns it into something you can see in one look, instead of a wall of text you read line by line.

13ms
for your brain to recognize an image
238wpm
how fast adults actually read prose

The same work, two ways

Same content, same facts. One you have to read; one you simply see. This is the whole idea.

rate-limiting-spec.mdread it
We need rate limiting on the public API before launch. Two approaches are on the table. A fixed-window counter is simplest: increment a per-key counter in Redis with a TTL equal to the window, and reject once the count passes the limit. It is cheap and easy to reason about, but it allows bursts at window boundaries, so a client can push close to twice the limit across a boundary. A sliding-window log is more accurate: store timestamps and count the ones inside the trailing window, which smooths bursts but costs more memory and a sorted-set operation per request. Given mostly steady traffic with occasional spikes, and the fact that we already run Redis, the recommendation is fixed-window for v1, revisited if boundary bursts bite. Open questions: what limit per key, and whether limits are per-token or per-IP.
The reviewer's job: decode it line by line and hold the whole thing in their head just to find the decision.
rate-limitingsee it
A request arrives
Count it in Redisone counter per key, set to expire
Under the limit?if not, refuse the request
ApproachBurst riskCostChoice
Fixed windowAt edgesCheapShip first
Sliding logSmoothedHigherLater
Chosen: fixed window2 open questions
The reviewer's job: none. The decision, the trade-off, and what is still unresolved are all visible at once.

What you can see

Anything an AI writes for you, turned into something you can follow at a glance. A few of the shapes it takes:

Specs

Goal, decisions, steps, and open questions, laid out instead of buried in a wall of notes.
Goal
Steps
Open questions

Roadmaps

What is done now, what comes next, and what is for later, as columns you scan in a second.
now
later

Feature comparisons

Options and competitors as a table, so the winner is obvious at a look.
OptionAB
FastYesNo
CheapYesYes
See a real one. This roadmap is a live page built and published with regrok, the same way you would make yours.
View the regrok roadmap →

How it works

Five steps take you from a wall of AI-written text to a page anyone can see and sign off on.

1

Capture

Paste whatever an AI wrote for you: a spec, a plan, a diff, a summary, and get a working page back.

2

Shape

Lay it out as goal, decisions, and open questions, then look up real examples of the same thing.

3

See

Turn it into flow diagrams, comparison tables, and status labels you can scan.

4

Publish

Push it to your subdomain with a link preview and a shareable image.

5

Review

Collect thumbs-up and comments right on the page. Undo any step.

How regrok is different

Plenty of tools now turn a prompt into a picture: type a prompt, get a polished deck, diagram, or UI you tweak in place. regrok works the same way, but four things make it different.

Gamma Napkin AI Beautiful.ai v0 by Vercel tldraw Make Real Claude Artifacts
01

It runs inside your AI session

Gamma, Beautiful.ai, and v0 are separate apps you have to leave your work to visit. regrok is a tool inside your AI session, so what the AI wrote never leaves the place it was written.

02

It starts from your real work, not a prompt

The deck tools build slides from a blank prompt. regrok reshapes what an AI just wrote for you: the spec, the roadmap, the diff, so the content is already yours.

03

The output is a page to read, not slides or app code

v0 and Make Real hand back running UI code; Gamma and Beautiful.ai hand back slides. regrok publishes a web page you share with a link.

04

Research and fact-checking are in the loop

expound adds real examples with links to the source, and fact_check flags shaky numbers. The others just format what you give them; they do not check it.

It runs in your terminal

regrok is a Model Context Protocol (MCP) server, the standard way tools plug into an AI assistant. Generate a token in Settings under Connect to Claude, run the command below once, and the whole toolkit appears as slash commands inside Claude Code. It all happens inside your coding session.

copy and run
claude mcp add --scope user --transport http regrok https://mcp.regrok.info/mcp --header "Authorization: Bearer rgk_..."
1
command connects the whole toolkit
0
context switches: you stay in Claude Code

The MCP toolkit

Connecting adds a full set of Model Context Protocol tools (and matching slash commands) to Claude Code, each scoped to your account by your token.

Create
start_from_textPaste a spec or a rough write-up and get back a sharpened, colored, published page in one call.
Shape and see
understandLay it out as goal, decisions, open questions, and steps.
sharpenRewrite into a visual layout of inline flow diagrams and tables.
expoundLook up real examples of the same thing and add them with links to the source.
fact_checkFlag shaky claims and numbers; repair applies the fixes.
reduce_verbiageTighten the wording without dropping any facts.
recolorSwitch the page to a different color palette.
Publish and share
publishPublish to your subdomain. Temporary by default; make it permanent to keep it.
pngRender a published page to a shareable image (a PNG file).
unpublishTake a page offline. The draft is kept.
deleteDelete a page, both draft and published copy.
Read and review
listList your pages, published and draft-only.
readRead the HTML source of a page.
updateReplace a page's HTML and republish it.
undo / redoStep a page back or forward through its last 5 versions.
feedbackSee the thumbs-up count and comments on a published page.

Frequently asked questions

What does “regrok” mean?

To grok something is to understand it fully (the word is from a 1961 novel). To re-grok is to grok it again: to understand, at a glance, the wall of text an AI just wrote. That is the whole idea. An AI writes a spec, a plan, or a diff in seconds, all as long text, and regrok turns it into something you can see in one look and act on, instead of reading every line. You capture what the AI wrote, shape it, and publish it to your own subdomain, all from your terminal.

What do you mean by "see it, don't read it"?

A spec written as prose makes a reviewer read it line by line and hold the whole thing in their head. The same content as a flow diagram, a comparison table, and a few status labels lands in one look. regrok does that for you.

Does it work with what my coding agent writes?

Yes. Most AI coding assistants can draft their whole approach before touching your code: the goal, the files they will change, the steps, the trade-offs. That write-up normally scrolls off in the terminal. regrok turns it into a clean page you can share and get sign-off on. It works just as well with a spec or notes you typed yourself.

How do I get access?

regrok is invite-only while in early access. Request an invite below and you will get an email when one is ready.

Do published pages expire?

By design, yes. A page you publish is taken offline after 1 hour unless you make it permanent. Your draft is always kept, so nothing is lost, and you can make it permanent or republish at any time.

Request an invite

Tell us who you are and what you would like to see instead of read. You will get an email when an invite is ready.

© 2026 regrok.info
Use cases FAQ Log in Request invite