GuideGen — Step-by-step guide capture

Record a workflow and auto-generate an annotated step-by-step guide. Export to HTML, Markdown, PDF, PowerPoint, and narrated video.

Do the workflow once. Get a narrated video that explains it. Turn GuideGen on, click through the thing you want to explain, and press stop. You get a step-by-step guide: one annotated screenshot per click, the target ringed and numbered, and a written instruction for every step — "Click Rider Management", "Type Demo in the Search field", "Press Enter". Edit any of it, then send it out as a video, a link, a file, or as text for an assistant. Same chore, different desks: a support lead writing a runbook, a QA engineer filing a reproduction, an ops manager handing a process to a new hire, a founder explaining an internal tool for the fifth time. 1. A NARRATED VIDEO, RENDERED ON YOUR OWN MACHINE • One slide per step at 1080p: the screenshot on a card, the click ringed and numbered, a step counter and a progress bar • A voice reads each step aloud. It is Backpocket AI, and it speaks on your own machine — the text of your internal processes is never sent to a speech service • Five speaking paces, from Very fast to Very slow. The pace sets the speaking rate and each slide's length together, so the narration never walks off the picture it belongs to • Slides crossfade and push gently toward the part of the screen you clicked • It comes out as a .webm file. The pictures, the voice and the muxing all happen in your browser 2. CAPTURE THE LAST 2 MINUTES Nobody knows in advance that they are about to want a guide. You finish the thing, and then somebody asks how you did it. • Switch catch-up capture on for one site — the admin panel, dashboard or tool you actually document • Then just work. A small dot in the corner is the only sign it is on; hover it to see how much is held • When somebody asks, press "Capture last 2 minutes" and what you already did becomes a guide, screenshots and all • Nothing can screenshot the past. On a site you have armed, GuideGen captures as you click and throws it away, keeping only a rolling window • Which is why it is off everywhere until you turn it on, one site at a time. Never in a private window. With a password field focused it keeps the written step and takes no picture at all • What is held never leaves your device and is deleted after 7 days whether you use it or not. Your dashboard lists what is waiting and how long each one has left 3. PUBLISH ONE GUIDE AND SHARE A LINK • Publish a guide, deliberately, and get a link anyone can open — never a background sync of everything you have ever recorded • The page carries your name, how long the recording took and the app it happened in, then the numbered steps. Any screenshot opens full-screen and zooms, because a dashboard shrunk to fit a column is a picture nobody can read • Any single step can be linked to on its own, so "look at step 7" is a link rather than an instruction • Pasted into a chat, the preview shows the guide's own title — never a step, never a screenshot. Shared guides are marked no-index, so an internal process does not turn up in a search engine • Re-publish and the page updates in place, so the link you already sent is never stale. Unpublish and the link stops working and the images are deleted • You can also let readers build their own web page, Markdown, PDF, slides or video from the page. They sign in to do it and you are shown who exported what, including their email address. It decides whether a button appears; it is not a lock on reading the page, and we would rather say so than imply otherwise 4. HAND THE WORKFLOW TO AN ASSISTANT • Copy for AI puts the whole workflow on your clipboard as plain Markdown: every step, the action it was, and the URL and page title it happened on, grouped by page • It is the written record rather than the screenshots, so you can read exactly what you are about to send. Paste it into any assistant, an IDE, a ticket or a note. There is no integration to configure and no API key • Pasting a handoff into an assistant does send it to that assistant. That is the point, and no tool can change it. What changes is that you decide, and you can see what you are sending first • The exports built for people leave that context out. A model needs it, or it guesses • GuideGen documents a workflow. It does not replay or automate one, and the handoff is text, not a script 5. WHAT THE PAGE ACTUALLY DID • Every recording keeps a summary of the requests each step fired and the status that came back — POST /api/orders → 500. It travels with the AI handoff and with no other export • Separately, and off by default, you can opt in to record a failed exchange in full and copy it as a cURL: the header names, the body sent, the body returned. "I clicked Save and it did not work" turns into something an engineer can start on. It is capped, truncated, and limited to responses of 400 and above • Credential header values, query-string values and obvious secrets in a body are replaced with a mask before anything is stored, and no setting keeps a real one. The cURL documents the call; it does not replay it • A step where everything succeeded shows one muted line. A step that failed shows the failure straight away, because that is the case you opened the guide for • Publishing a guide never publishes its log 6. SIX WAYS OUT OF ONE RECORDING • AI handoff — the whole workflow as Markdown on your clipboard • Web page (.html) — one self-contained file • Markdown (.md) — images embedded, ready for a wiki • PDF — title page plus paginated steps • Slides (.pptx) — one slide per step • Narrated video (.webm) — the walkthrough above Every one of them is built inside your browser. 7. MADE TO BE READ • Step text comes from the control you actually clicked — its label and its role — not a generic "click here". Fields, dropdowns, checkboxes and keystrokes each read the way a person writes them • Screenshots crop to the part of the page that matters instead of shrinking a whole window until nobody can read it • Tab switches and page loads are recorded as their own steps, so a workflow that opens a new tab still reads in order — and a page load that only restates the click before it is left out • Typing is one step per field, with the screenshot taken once you have stopped • A title is suggested from the recording, and you can rename it in one field • Light, dark or follow-your-system, everywhere — the popup, the editor and the guides you send • In the editor: rewrite, reorder or delete any step; add steps of your own between the recorded ones, with a picture or without — without one, the text becomes a section slide in your exports; and drag a box over anything sensitive to pixelate it in every export, permanently 8. IT STAYS ON YOUR MACHINE UNTIL YOU PUBLISH Guides are saved in your own browser. Nothing is uploaded in the background, for processing, or on a schedule — only the single guide you press Publish on. Annotations and redactions are burned into the image before it leaves your machine, so no unredacted original is ever sent, and password fields are masked the moment they are captured. That is what makes GuideGen usable where cloud tools cannot go: regulated work, real customer data, internal admin panels, anything under an NDA. GOOD TO KNOW • A free account is needed. It signs you in to the editor and is what makes publishing work • The editor is a page on guidegen.backpocket.website. It reads the guides held in your browser over a channel only that site can use, so the editing happens there while the guides stay here • Works on web pages in Chrome and other Chromium browsers (Edge, Brave, Arc). It cannot capture native desktop apps — browser only • Guides you have not published live on the machine that recorded them, so they do not appear on your other computers. Published ones do • The download is larger than most extensions, because the narration voice ships inside it rather than being fetched later • The first narrated video takes a moment longer while the voice loads • If a tab was already open when you installed GuideGen, reload it once before recording • It documents a workflow; it does not replay or automate one. If you need a script, a browser automation recorder is the right tool • Step text is written in English. It will record an interface in any language, but the sentences around it are English Free, and no card.