Autocapture agent: build a template end-to-end from a prompt, link, or source file
Use the Autocapture agent in Walnut or the Chrome extension to turn a prompt, product link, recording, PDF or existing demo into a captured template, then steer or stop the run.
Autocapture agent: build a template end-to-end from a prompt, link, or source file
About this featureβ
The Autocapture agent turns a short prompt into a full template. Point it at a live product URL, paste a link to an existing demo (Walnut's or another tool's) or a screen recording, or attach a video or PDF deck. The agent plans the flow step by step, drives the tab for you, captures each screen as it goes, and files everything into the destination you chose.
You can start it in two places: from New β Template in Walnut, or from Agent mode in the Walnut Chrome extension. Either way, the extension does the capturing on your live product.
You can edit the plan before it runs, steer it mid-capture, and stop it at any time without losing the screens it has already kept.
Snapshot πΈβ
- Where it lives: in Walnut, My Library β New β Template (an empty template shows the same page). In the Walnut Chrome extension, Agent mode in the popover and in Chrome's docked side panel.
- Requires: AI access turned on by an admin (Company Settings β Account Settings β AI), a role that can capture screens (Admin or Editor), and the latest Walnut extension (v2.7 or later).
- Captures into: a new template, an existing template, or the Screens Library in the Resource hub.
- Starts from: a description, a product page, a video or PDF file, a link to a video (including Loom and Vimeo), a link to a page, or a link to an existing demo built in Walnut or another tool.
- AI credits: runs use AI credits. They appear as Autonomous capture agent in your AI usage.
Start in Walnutβ
- Go to My Library, click New, and select Template. (Opening a template that has no screens shows the same page.)
- Under What demo do you have in mind?, describe the demo you want, or pick an example such as Full product tour.
- To work from something you already have, click Start from something you have. You can upload a recording (MP4, MOV or WebM, up to 1 GB) or a written plan as a PDF (up to 50 MB), or paste a link to a video, another demo or a page about the flow. Export a PowerPoint deck as a PDF first.
- If you don't say what to learn the demo from, the agent asks. Pick Upload a recording or written plan, Share a demo link, Give me your product URL, or Use my description.
- Given your product's URL, the agent can look through a few pages in a background tab and suggest the flows worth showing. Pick the ones to cover, or click Skip the scan. Sign in to your product in the same browser first, so the scan sees the product and not a sign-in page.
- Review the draft. Rename the template, click Edit steps to add, reorder or delete steps, or type what to change. Each step is marked Seen in product or Inferred. Steps that reuse screens from your library are marked Reused, added first.
- Click Create template, then confirm where the capture starts (Start here or Use a different URL).
- The extension opens your product in a capture tab and runs the plan. The chat shows Capturing⦠N of M screens. If the agent has a question, click Go to capture tab to answer it.
- When the run ends, click Open template. If the agent couldn't reach a planned step, the chat names it. The template is saved either way, so you can add the missing screens yourself.
Prefer to build it yourself? Under Or build it yourself, choose Capture new screens, Import from Library, or Import from Figma.
Start in the Chrome extensionβ
-
Open the Walnut extension on the tab you want to capture, and switch the extension to Agent mode.
-
Under Capture to, pick where the screens will land: New template, an existing template from the dropdown, or the Resource hub (optionally choose a product and feature label; every screen the agent captures gets them).
-
Type what you want to capture into the composer. You can:
- describe the flow in plain language ("Capture a buyer going through sign-up and creating their first project"),
- paste a link to a video (Loom, Vimeo, or any https recording), a public page, or a demo built in Walnut or another tool, or
- use the paperclip to attach a video or PDF deck.
You can send just a link or a file with nothing typed. The agent then asks what the demo is for.
-
Press Send prompt. The agent asks a quick question about the audience; pick an option or Skip.
-
The agent drafts a story β the beats it plans to cover β and then a plan of captured steps. The status badge shows where it is: Researching, Story ready, Planning, Plan ready.
-
Edit any step title inline, drag steps to reorder, or type a change into the composer and press Re-plan to redraft.
-
When the plan looks right, press Start capture Β· N steps. If the agent needs a different start page, it opens it in the tab first; if a sign-in is required, finish signing in and the agent carries on.
Starting from a link, a video or a deckβ
- A product page link in your prompt is picked up as the start page for the capture.
- A video link (for example a Loom or Vimeo recording of the flow) or an attached video file is read by the agent, which drafts the plan from the recording. If the link only plays when the tab is visible, the agent offers Open and play or Attach the recording.
- A PDF deck is read the same way β each slide becomes context the agent uses to plan the steps.
- A link to an existing Walnut demo starts the capture at the demo itself; the agent walks through the demo and lifts its steps into the destination template as a guide.
- A link to a demo built in another tool (for example Storylane) works the same way. The vendor's own guide boxes, spotlights and hotspots are left out of the captured screens.
From a recording, deck or link, the agent first drafts the story the source tells and picks the screens that carry it, leaving out detours. If you want a screen it left out, add it back when you review the story or the plan.
Follow the runβ
Once the agent starts, the capture panel shows up on the page. You can keep it as a floating window over the page, dock it to Chrome's side panel, or minimize it to a draggable chip and tap the chip to restore it.
The panel header shows:
- the flow name and a status badge (Needs you, Live, Finished, β¦),
- a progress bar with N of M screens captured, and
- a Stop capture button.
Each planned step shows one of:
- Live β the agent is working on this step.
- Captured β the screen has been accepted and filed.
- Waiting β the agent is asking you a question (see below).
- Missed β the agent could not reach the step.
- Up next / Planned / Not reached β not started yet.
If the last screen was not kept, the panel prints the reason beneath the live step so you can decide whether to steer or stop.
Activity at the bottom expands a turn-by-turn log of what the agent is looking at and the verdict on each screen it has tried.
Answering a question mid-runβ
The agent can pause and ask you something β which of two options to pick, where to click when a control is ambiguous, or whether a screen is the right one. The question shows up on the step that is waiting, with the choices as chips; typing a free-form answer in the composer sends it as the reply.
Adding a noteβ
While the agent is working you can send a note through the composer. Notes don't interrupt the run β the agent reads them on its next turn and uses them to steer the rest of the capture.
Stoppingβ
Stop capture ends the run cleanly. The current turn finishes and any screens already accepted are kept and filed; nothing is discarded. The button stays in a Stopping⦠state until the panel confirms the run is over.
When the run finishesβ
- A template capture files each accepted screen into the chosen template in the order the agent walked them. The agent also links each screen to the next from the click it made, so check the links in Preview.
- If you have AI Mode, open the template and AI Mode asks what to make of the capture: Guides, Sandbox, or Hybrid, with its suggestion. Pick one and keep editing in chat. See AI Mode.
- A demo capture (started from a link to an existing Walnut demo) writes the demo's steps as a guide onto the captured screens.
- A Resource hub capture lands every screen in the Screens Library under the product and feature labels you picked, ready to reuse across templates.
The panel's Summary view at the end of the run has a shortcut to open the destination.
FAQsβ
Can I use the Autocapture agent on an app that needs me to log in?
Yes. Sign in on the tab first, then start the agent β it carries on from the signed-in page. If the agent lands on a sign-in page it cannot pass on its own, it waits and prompts you to sign in and continue.
Can I keep using the tab while the agent runs?
The agent drives the active tab, so let it work there. Dock the panel to Chrome's side panel if you want a stable place to follow the run while the tab navigates.
What happens if I close the popover during a capture?
The run keeps going. Reopen the extension or the docked side panel to follow it again. Closing the popover does not stop the run; only Stop capture does.
What if the plan isn't quite right?
Edit step titles directly in the plan card, or type a change into the composer before you start β for example "skip the settings step" or "add a step for inviting a teammate" β and the agent re-plans. You can also press Start over in the header to clear the thread and begin again.
Why don't I see Agent mode in the extension?
Agent mode appears only when an admin has turned on AI access for your company, and only for roles that can capture screens. Update the extension and reload the page if it's still missing.
Does it work with any browser?
The Autocapture agent is part of the Walnut Chrome extension and requires extension v2.7 or later. Update the extension if the agent option is missing from your popover.