Auto Capture films a screen walkthrough of a web page with no one at the keyboard. Ask the Tella Agent in the editor, or Claude, ChatGPT or another AI app connected to Tella’s MCP server.
Auto Capture is in a closed beta and isn’t available to every workspace yet. We’d love your feedback: email support@tella.com.
Auto Capture films a screen walkthrough of a web page with no one at the keyboard: an AI agent opens the page in its own browser, explores it, plans the demo, and records it with an animated cursor. You start it by asking the Tella Agent in the editor, or an AI app such as Claude or ChatGPT connected to the Tella MCP server.Use it for a product demo, a feature tour, a how-to clip, or b-roll of the page you talk about in a recording, when you don’t have a screen recording of your own.
Each request is called a take, and one take films one page or flow:
Explores. The agent opens your URL in a fresh browser, looks around the page and decides which features to show and in what order. If the page needs a sign-in and you’ve saved a login for it, the agent signs in here, off camera.
Scripts and checks. It writes the walkthrough as a script and runs it to make sure every step works. Anything a check saves is put back first, so the recording starts from your account as it was.
Records. It plays the script back in a clean browser and films it, moving an animated cursor to each thing it points at.
Delivers. The finished take is processed like an upload and handed back with a short account of what it shows, chapter markers, and suggested cuts.
A take usually runs for several minutes. Part of that can be waiting for a recording machine to start.
The agent starts the take and waits for it. Keep editing while it runs.
When the take is delivered, the agent looks at it and places it: as b-roll over part of your video, or as a clip of its own in the right spot. It also applies the suggested cuts it agrees with.
The agent can also start a take on its own when an edit needs footage of a web page your video doesn’t have, like a step your recording skips. It tells you in one sentence what it’s recording and why before it starts.
In a chat with the Tella connector turned on, ask for a recording of a page, with its full URL: “Use Tella to record a product demo of https://example.com for new visitors.”
The app starts the take and checks on it every 30 to 60 seconds until it finishes. Some apps only check when you send another message: if yours goes quiet, ask “Is the Tella capture done?”
When the take is delivered, ask the app to make it a video (“Put it in a new Tella video”) or to add it to an existing one. The take isn’t in your library until it’s placed in a video.
Open the video in Tella to edit, trim, add narration or share it.
For a longer video that covers several pages, ask for one take per page or flow, then have them added to the same video in order.
Describe the outcome you want rather than a list of clicks. The capture agent explores the page and finds the buttons itself:
Who it’s for and what it’s for: “a product demo for first-time visitors”, “a how-to for admins setting up SSO”.
Anything you specifically want shown, skipped or ended on: “compare the three plans and stop on the yearly toggle”, “don’t open the settings menu”. Steps you name are followed; anything you leave open, the agent decides.
A length, only if you need one, such as “about 20 seconds to fill the gap at 1:20”. Without one, the take is as long as the demo needs.
Whether to show how to get there. The agent can navigate to a feature off camera and start the recording on it. If viewers need to see the path, say so.
What may change on the account, if the demo saves something: whether it’s a sandbox, which record to use, how to undo a change, and what mustn’t change, such as “don’t send anything”. If you don’t say, the Tella Agent or your AI app should ask you before a take that saves something. Either way, use a demo account you’re happy to have clicked around in.
Keyboard shortcuts, if you want them. By default the agent clicks every control, so viewers can follow it. Ask for a keyboard-driven demo to have it use shortcuts where someone who knows the app would, then turn on Keystrokes so viewers see them.
Auto Capture uses the signed-in account for real. Anything the demo saves is saved while the agent explores, on every check, and once more in the recording.
Changes it can undo: before each run, the agent puts back what earlier runs saved, such as deleting a project an earlier run created or turning a setting back off. Every run, the recording included, starts from the account as you described it.
Changes it can’t undo: for something like sending an email or confirming an order, the agent stops before making the change and asks you what to do. It can’t yet keep such a change for the recording alone: every check makes it too. So answer with a way round it, for example “use a test account where that’s fine to repeat” or “stop on the confirmation, don’t send”, and a new take starts with your answer.
What’s left changed: when a delivered take leaves something changed on the account, such as a new record it created, your AI app or the Tella Agent tells you.
Auto Capture records any http or https page on a public domain. It can’t record:
Pages on your own computer or private network, such as localhost, IP addresses, or intranet hosts.
URLs with a username or password in them.
Desktop apps, mobile apps, or anything outside a web browser.
Captures run on Tella’s servers, not on your computer, so they see a page the way a new visitor does. Some sites block automated browsers or traffic from cloud servers and show a “verify you’re human” or “sign in to confirm you’re not a bot” page instead; Auto Capture can’t get past those. If it’s your own site, you can let captures through with a firewall rule: see Let Auto Capture past your firewall or bot protection.
To film signed-in parts of your product, save a login for it first under Settings → Credentials. See Let Auto Capture sign in to your product.When a saved login matches the page, the agent signs in once while exploring, off camera, and the recording starts already signed in, with no login screen in it. The sign-in is filmed only when you ask to show it, or when the site’s sign-in can’t be carried over to the recording browser. Password fields are always masked in the recording.If the page needs a sign-in and you haven’t saved a login for it, the take stops and you get a link to Settings → Credentials. Add the login there, then ask for the take again.
Never paste a password, code, or sign-in link into the chat. Auto Capture only signs in with logins saved in Tella, and asks for codes on a Tella page.
If your product sends a code by email or text message when the agent signs in, the take pauses and asks you for it:
Your AI app or the Tella Agent gives you a link to a page on Tella. Some apps open it for you.
On the page, check the site under Tella would like to sign in to, and the Account when it’s signing in with a saved login.
Type the code you received into Enter code and click Allow access.
When you see You’re all set, close the page. The take carries on by itself.
The page shows how long is left. If the time runs out, the take stops and you’ll need to ask for it again. Only the person who started the take can see or answer the request. Saved logins that are yours (Only you) can ask for a code; workspace logins can’t.A take can also ask you for a code with no saved login, when a site signs you up or in with only an email address and a code it sends there. Give the email address in your request, such as “sign up with demo@example.com”. The code is asked for only while the agent explores, so the recording starts signed in, after the sign-up.To skip this step, save the login from 1Password with its one-time password, and Tella fills in the code itself. See Sign in with two-factor authentication.
What you get when an Auto Capture take is delivered
Along with the recording itself, each delivered take comes with:
A handoff note: what the clip shows, anything the agent did off camera that isn’t in the clip, and where it departed from your request.
Chapters: where each section of the walkthrough starts, useful for titles, narration or splitting it up.
Suggested cuts: waits for pages to load and over-long pauses that can go without losing anything. The Tella Agent applies the ones it agrees with; in other AI apps, ask for them to be applied.
Key presses: every key the take pressed is recorded with it, except what’s typed from a saved login, so you can show them with Keystrokes.
The take arrives as a screen recording with an animated cursor, so you can edit it like any other clip: trim it, add zooms, restyle the cursor, add a voiceover, or change the background.
If a delivered take is almost right, ask the Tella Agent or your AI app to revise it and say what to change: “Revise that take so it also clicks Completed and ends there.” The agent starts a new take from the script and notes the earlier one left behind. It changes what you asked for and keeps the rest, so a revision is usually quicker and uses fewer credits than starting over.
The earlier take is never changed. The revision arrives as a new take for you to place.
Only the person who started a take can revise it, from the same workspace.
A take that’s still running can’t be revised yet. Wait for it to finish first.
A take that stopped at a sign-in or on a page that blocked the capture browser can’t be revised. Ask for a new take instead.
Takes made before revisions were available can’t be revised. Ask for a new take with the change in your request.
To shorten or reorder what you already have, edit the clip in the editor.
Ask the Tella Agent or your AI app to cancel it, for example after starting one with the wrong URL. A cancelled take usually stops within seconds and produces nothing to place.
A take ends without a recording in one of these ways, and your AI app or the Tella Agent tells you which:
It needs guidance. The agent got stuck on something only you can answer, such as which of two accounts to use, or reached a change to your account it can’t undo, and asks you a question. Answer it, and the Tella Agent or your AI app revises the take with your answer.
It needs a login. The page asked for a sign-in and no saved login matched. Add one under Settings → Credentials, then try again.
It failed. The message says why. If the failure was temporary, like the AI model being briefly unavailable, ask for the same take again in a few minutes. Otherwise, check the URL works in a private browser window.
Auto Capture is included with paid plans and uses your workspace’s credits, the same balance the Tella Agent spends. Each take uses credits for the AI that explores and scripts the page and for the machines that record it, and is charged when the take ends, including takes that end with a question, fail or are cancelled partway. A take can’t start when the workspace is too low on credits; ask a workspace owner to add more.If the Tella Agent doesn’t offer to record pages and Credentials isn’t in your settings, Auto Capture isn’t available for your workspace yet.
Auto Capture runs on Anthropic’s Claude by default. To use a different model, or have it think harder on a complicated page, name it in your request: “use Claude Opus at high effort”. More effort takes longer and uses more credits. Most takes don’t need either.
