New features and product updates.

Extension 38.3.13 is with the Chrome Web Store for review, and Chrome installs it on its own once it is approved. It carries the extension side of this release: reading pages without a tab, Map, Free Mode on paid plans, longer runs, lesson-site widgets, next steps and the one-click repeat schedule. Extension 38.3.12 shipped on November 5 with the act agent reading every open tab in full and the dropdown fix below. Nothing in either needs new permissions.

Screenshots and files come back as links from /agent. Every screenshot, image, PDF and file a Cloud run makes is listed in the response's artifacts, each with a link that works for 7 days, its size, a sha256 checksum, the page it came from and when it was captured. GET /artifacts/{id} returns the same list with fresh links. Add options.evidence.screenshots: true and the agent captures every page it takes data from and tags each record with the screenshot's evidenceArtifactId, so each row can be checked against the page. Webhooks list images too, and MCP cloud_agent returns the same list.

See where the managed proxy can exit before you run. GET /proxy-locations lists every country code a location accepts and whether the managed proxy had an exit there when we last checked: 120 of 249 countries on November 11. It also gives the region and city format and the fallback a run uses (city, then region, then the whole country, never another country). The docs list the countries, and POST /proxy-location/check still answers for a region, a city or a list right now.

Pages read without opening a tab. The agent can now read a page by its URL off the browser: directly first, then through the residential proxy, then through an unblocker on a bot wall (0.15 credits a page). It gets the same page structure a tab read gives, and an error page counts as a failed step instead of content. This is live in Cloud and over MCP fetch now, and reaches the extension with 38.3.13.

Map lists a site's pages before reading them. Map reads the site's robots.txt and sitemaps and, when they list nothing you asked for, follows the site's own links for a short crawl, with no browser and no model. A section such as docs.python.org/3/library/ that the sitemap leaves out now returns its 307 pages instead of none. It is free, apart from 0.15 credits a page when a bot-walled page needs the unblocker (at most 3 a map). Use it from the API (POST /map), MCP map and Cloud tasks now, and in the extension with 38.3.13.

Every reasoning level in Free Mode on any paid plan. A paid plan runs Free Mode's value models at every reasoning level, up to Max, while premium models run on credits. Schedules use the daily Free Mode allowance first, then credits, and each schedule has a Billing choice: Free Mode first, or credits only with no ads. Cloud has this now; the extension gets it with 38.3.13.

Next steps after every finished run. A completed run suggests what to do next, its first suggestion fills the empty message box, and Tab sends it in. A short reply such as "next" or "the second one" now goes out as the suggestion it names, instead of starting the work over. A new user's first finished run also offers to run it again tomorrow.

One-click schedules for tasks you repeat. When you run the same prompt on two or more earlier days, the side panel offers to schedule it on that Chrome, emailed to you by default. With 38.3.13.

Long runs keep going. A pass no longer starts with under 3 minutes left in a run's 45-minute window, the window renews while the run keeps making progress, and the time a run spends waiting for your answer no longer counts against it. With 38.3.13.

Runs check their own work. The end-of-run review now reads a record of every call each script actually made, so a run's last pass answers from what happened rather than from the script's own summary. It runs once per run and no longer uses up the run's repairs. In Cloud and in the extension.

Longer tasks stay quick and cheap. A run keeps every earlier step's element ids and history until a size budget, instead of dropping the oldest after 20 steps and re-sending the rest, so more of each pass is read from the model's cache. Since November 5 the act agent also reads every open tab in full (38.3.12), and a READ ONLY tab needs no switching.

GPT-6 Luna, GPT-6.1 Sol and GPT-6.1 Terra run on Bedrock global first, then regional Bedrock, then OpenAI, on every round. The models page shows the order.

Lesson and course sites. The agent picks a radio answer by its label, drags with the keyboard where a widget only takes keys, describes canvas images, types into spreadsheet-style grids and commits with Enter, plays a lesson video the page waits on, and answers listening questions from the page's audio instead of guessing. Icon-only buttons are named by their icon, so the right one gets clicked. With 38.3.13; the radio, drag, canvas and grid fixes also run in Cloud.

Lists extracted from one page stay whole in Cloud. Pulling a page's items into a structured result (all the quotes on a page, say) no longer comes back split into one record per item, which had left a saved CSV with 1 of 10 rows.

Dropdowns and network reads. On a form of token pickers, another field's chosen chip no longer counts as this field's answer (38.3.12, and in Cloud). A network extract stops at the timeout it asked for, with a 60-second minimum, instead of 45 seconds later.

Model failovers. A DeepSeek Flash image-count error moves the call to the next supplier at no cost. In the extension, an image-budget error fails over to OpenAI and AVIF images are converted to JPEG instead of dropped. With your own Anthropic key, MCP tools with dots in their names work instead of falling back to managed models and billing credits.

Cloud and Tailscale. Asking Cloud to open localhost or a private address gets a plain refusal instead of a server error. A run paused for a Tailscale sign-in hands its tunnel to a newer run instead of holding your one route for 10 minutes, and auth-key refusals say they are auth-key problems.

For AI apps and the API: Enable Extension as MCP is on by default on every plan with 38.3.13, free tools are listed first, and a tool your plan cannot use answers with next steps instead of a protocol error. MCP gains fetch and map, and the API gains POST /map, GET /proxy-locations and GET /artifacts/{id}.

Smaller fixes. The Free Mode device-limit notice names the masked account that holds Free Mode on that computer, with a Switch account button. The post-install page opens Retriever with a first task ready (38.3.13). Gmail's connect window explains its consent screen, Confirm cancel stays readable on hover, and the /scrape docs describe the page tree as the text outline it is.

Extension 38.3.11 is rolling out on the Chrome Web Store, and Chrome installs it on its own. It carries the extension side of this release: the form, dropdown, click and drag fixes, embedded frames, long waits, the new side panel message with one export button and a Schedule button, paste to attach, list batches and deep search. Extensions 38.3.9 and 38.3.10 shipped since October 31 with sheet status tabs, steadier background reads and the Google Sheet tab fix. Nothing in it needs new permissions.

Web lookups answer from the source's own page, in fewer passes. To look up a fact such as a version, a date, a price or who holds a role, the searching script now also opens the result on the source's own site, the next pass answers from that page and cites it, and "latest" means what the official page says. In our Cloud tests on four lookups, runs went from 4.9 planner passes and 5.9 searches on average to 3.2 and 1.8, every answer cited the official site, and model credits per run fell from 0.59 to 0.42. Search results count as snippets, not the answer, so a run no longer ends "success" with the field it searched for left blank. A local search the script did not flag ("near me", a service in a city) now gets places results first, and a name with "linkedin" gets a LinkedIn-filtered search. In Cloud and in the extension.

Forms and dropdowns fill in one call. Typing into a choice field now picks the option the text names and reads it back. select_dropdown_value picks only from the field's own list (it had clicked options in a neighbouring list or the site's menu), checks the field's live value, types into the popup's own search box, gives Workday prompts their Enter and matches across accents and word order; a native select that lacks the exact wording names the nearest options, such as "BS" for "Bachelor's". The read after a click waits for the page to settle, so a form is no longer caught mid-submit, and it leaves open the menu the click opened, LinkedIn's More included. In Cloud and in the extension; needs extension 38.3.11 in the extension.

Helper steps think less, and Cloud plans at full speed again. Since November 1, act steps, page extraction and text processing run one reasoning level below your setting while the planner keeps it: max to high, high to medium, medium to low, low to off. DeepSeek, GLM and GPT-6.1 Sol stop at low, and an Off you picked stays Off. In our tests Luna act steps scored 98.3% at off, low and medium alike and page extraction 100% at every level, and image captions now really run at low: 2.3 seconds at the median instead of 3.9. Cloud's planner had been writing about 40% slower than the extension's since October 30 (a median planning call went from 8.1 to 13.2 seconds for the same output); this release fixes the routing that caused it. A planning pass over about 250K tokens now sends only its last three results in full, so very long runs stay quick. In Cloud and in the extension.

Tables, lists and results come back whole. A table you ask for now finishes as rows under a named field, never a Markdown string, and the Cloud result card also reads a table returned as text into rows: a "return them all as a table" run that showed 3 records now shows all 30, with Table, Sheet and CSV working. Opening each item of a list dedupes its links and reads the fields you named with extract: an 8-book task that ended Partial with its price column empty now returns all 8 rows with every field and no duplicate tabs. The finished result holds the answer fields and a summary that leads with the answer, never the page reads that checked them. A follow-up in the same chat treats earlier replies as finished tasks and checks their facts again when the new task depends on them. In Cloud and in the extension; the extension's Result and MCP output need 38.3.11.

Plans take fewer passes and ask less. The planner now puts the page steps it can already name in one script and hands a run of dependent steps to one act, where a third of planner passes had only clicked or typed one to three times and read the page back. An act step the page confirmed counts as done, so acts stop answering "could not be verified" after doing the work. An act now takes up to 25 steps by default, up from 15: 87% of acts that stopped at exactly 15 were followed by another act averaging 9.8 more steps. In the extension, the planner asks first only for a required input nothing supplies or an outward step you didn't ask for, and a "don't ask" holds for the rest of the run, never for a sign-in, payment or personal detail. In Cloud and in the extension; the 25-step default and "don't ask" need extension 38.3.11 in the extension.

PDFs fit the page count you ask for. A one-page or landscape request now reaches the generator, which steps type, spacing and margins down until the PDF fits without cutting anything and states the page count, so the run no longer reads it back. Long table cells stay whole, and a row taller than a page carries on under the repeated header. In our test, a one-page PDF of every quote on a tag, which had run the generator four times and paused at eight steps, came back from one call as one page with all 14 quotes and authors, in 1.5 minutes and 2.72 credits. A file that still misses a layout after one retry is delivered with what differs, never generated a third time. In Cloud and in the extension.

Deep web search, and research on the free plan. The planner can now run one deep search that reads the top pages and writes a cited answer, in the shape you ask for when you name fields: about 10 to 12 seconds, billed as 8 searches (free units first, then 1.2 credits a call), and nothing when it fails. On the free plan, research now runs in Free Mode runs, paid only from your free monthly data-tool allowance and never from credits. Research always runs at its low effort, about 2.5 credits a call on paid plans, and now asks for the number of rows you want: 64 of 65 Cloud research calls had left it out and stopped near 10. Deep search is in Cloud now and in the extension with 38.3.11 and Web Search on; the research changes apply in both.

Runs wait out long jobs. A wait can now last its whole duration, up to 2 minutes at a time, so a render, an app build or a queued video is waited out instead of ending at the page's first change, and the planner waits out a running job in one script, up to 10 minutes on the finished control. Before, extension runs spent 769 planner passes in two weeks on 10-second waits that ended early (AI Studio builds took 63 to 67 per task). Stop still ends a wait at once. In Cloud and in the extension; needs extension 38.3.11 in the extension.

A cleaner side panel answer, with one export and a Schedule button. A finished run's message leads with the planner's reasoning, then one closed row per step, then Artifacts and the Result, one level deep; Copy still gives the whole run, and a data block longer than 12 lines folds to its first 6. A table gets one export button: Google Sheet when Sheets access is granted, CSV otherwise. Every completed run has a Run this on a schedule button, prefilled. Pasting a skill link, @name or skill name attaches the skill, a Google Sheets link joins the run's tabs, and clipboard files upload as if dropped; the Skills menu leads with Suggested and Recent. Guidance you type stops a running script so the next pass plans with it, and sent guidance shows ✓, then ✓✓ once read. A one-question choice answers on the first click, a retry with a note carries the task, and Connect Gmail sits at the top of the Tools menu. Needs extension 38.3.11.

Skills, schedule cards and Gmail in Cloud. A new Skills button in the Cloud composer opens a picker with search, Suggested, Recent and All skills, and the skills you pick ride the run; pasting a skill link, @name or skill name attaches it, a Google Sheets link attaches as a sheet even past the 3-link limit, and clipboard files attach like the Files button. Each skill in the Skills panel has Use in a run. A run that makes schedules shows a card for each one, and a repair pass now updates the run's schedule instead of making a duplicate. Connect Gmail is in the Tools picker and in Secure logins. Generated files are named by their title, never a storage path, and a run's read-back copy of its own PDF no longer shows as another file.

Run in my browser waits for your answer. A run sent from the website to your Chrome that stops on a question now shows the question, with no confetti, and your answer goes back to the same browser to carry on; before, it read Done, as 86 of 225 such runs in 30 days did. A retry carries the task, and a send that cannot start says why (no browser picked, offline, signed out, update needed) and offers Open Retriever, Check again or Run in Cloud. A template's Run button is no longer greyed out for a missing browser, an account with one connected browser has it picked, and Cloud's sign-in wall offers Run in my browser when the extension is connected.

Sign-in codes from your inbox, and Gmail sends that report right. When your task asks for it, the agent can type your email at a sign-in page and read the code from your connected Gmail, and with no Gmail connected the connect offer says it covers sign-in codes. A Gmail send or draft now returns its sent flag and its message, thread or draft id, where 35 real sends from October 19 to November 2 had been reported as failed. A missing or revoked Gmail permission says to reconnect, and the next connect asks for it again. In Cloud and in the extension.

Sheet jobs keep their place, on any sheet. When a task or skill limits which columns may change, a sheet batch keeps its statuses in its own "rtrvr status" tab keyed by your key column, so sorting or inserting rows keeps the progress and a repeated key runs once. Any task with a Sheets link or an attached spreadsheet now gets the sheet batch guidance (on October 31 all 11 runs that worked through a sheet's rows by hand had missed it), and so does a task that puts per-item results in a sheet. In the extension, a per-item job over a list with no sheet keeps its progress the same way. A Google Sheet the run reads no longer opens a second copy in front of your tab, and a new sheet given rows writes them. A view-only read of a sheet the run has no access to reads its page instead of asking first: 44.8 seconds in our Cloud test, where it had waited 11 minutes. In Cloud and in the extension; list batches need extension 38.3.11.

Executions show where your credits went. The Executions page has Today, This week and This billing cycle chips with totals for the period, a project filter for runs tagged from AI apps or the API, and now lists runs that AI apps started in your own Chrome, which were missing. The Account card shows the cycle's spend split into one-time and monthly credits, and get_current_credits reports the cycle's real spend instead of 0.

For AI apps and the API: agent, scrape, extract_from_tab, cloud_agent and cloud_scrape take a project tag that list_executions filters on; scrape, get_page_data and take_page_action point list gathering at extract_from_tab (one app had spent 45 calls and 88 minutes on 9 listings); a browser that stops responding answers not retryable, with a retry-after and, after 5 minutes, a note to open Chrome; a busy browser queues scrape like the page commands; on the free plan, a scrape or agent call with no browser to run in is told what it needs instead of going to the paid Cloud browser; check_results questions are answered on GLM Flash; and a client that keeps sending a dead API key gets 429 with Retry-After.

Background tabs and parallel runs read and click more reliably. A click in a tab you are not looking at that changes nothing is now always retried with the tab shown, instead of being refused after six shows in ten minutes (117 clicks from 20 users failed that way in the week before). At the task's tab limit, an act that wanted a new tab opens the page in its own tab instead of being refused, as 313 of 492 such opens had been inside parallel runs. Parallel reads take turns at the page, and a read that ran out of time gets one more try, where a quarter of reads in fan-outs of four or more had come back empty. Typing that timed out in a background tab brings the tab forward and checks the field before typing again, and activate_tab shows a tab until it draws, then gives you yours back. Since 38.3.9, a page that keeps changing in a background tab, like Google results, is read without a 3-second wait. Needs extension 38.3.11; the tab-limit and typing fixes also apply in Cloud.

Embedded frames and downloads. Frames now get their read time after the page itself is read, a child frame answers before its parent stops waiting, and frames identify themselves even in background tabs, so notebooks, course players and embedded apps read from inside: in our tests Colab read all 8 output frames in a background tab, and a lone 3,000-row frame read whole where it had come back with 302 to 2,039 rows. Ad frames no longer mark retail and game pages partial. A PDF shown in a viewer frame downloads as the PDF, not the viewer page; download_file saves any file except a web page, and an installer only when you ask for one; and a link that answered 401 to 410 is not fetched again in the run. Needs extension 38.3.11.

Clicks, drags and keys land. A click whose effect shows on the next read (its target changed, or its popup opened) now counts as done, and a step with an unconfirmed result reads the page again instead of re-sending the old view. Look-alike buttons, such as each row's Edit, are told apart by their row. A drag presses the widget's own hit layer and follows its pointer capture, keyboard handles drawn in graphs (Khan Academy's points, where all 19 drags had failed in a week) can be focused and moved, press_key takes a count of up to 20 arrow, Tab or Backspace presses in one call, a video slider lands within a step, and typing into a web terminal sends each key once. In Cloud and in the extension; needs extension 38.3.11 in the extension.

Cloud runs keep their work and their credits. A Cloud run that ends in error before its first step is no longer billed for browser time. A run whose time limit cuts a planning call after it already has output keeps that work: a run you are watching pauses with Continue or Stop here, and a scheduled one ends partial. Continuing a paused Free Mode run after a page reload no longer fails with "Could not verify Free Mode", which 14 of 23 such refusals had been. A write to a Google Sheet the run cannot open stops and says so instead of typing into the grid.

Smaller fixes. For you suggestions think at their usual level again, after the November 1 change had left about a third of feeds empty. A free-plan Chrome schedule saved to run more often than daily moves to daily once, with one notification, instead of being refused at every run, and the one-schedule rules also apply when you resume, switch on or edit one. A new Rover site no longer reads Live before Rover has loaded on it. Chat history no longer shows a Firebase Storage permission error on every panel open. A Google Doc or Slides deck that Drive refuses costs no credits and names the access it needs. A file excerpt a plan returns to itself shows whole up to the step's budget instead of 3,000 characters. Extensions older than 38.2.0 are asked to update before planning, with nothing run and no credits used. An act the model provider refused is not sent again, the network tool runs only when you want every record and says quickly when a site has no usable endpoint, and the website's contact links open Gmail compose in a new tab.

Extension 38.3.8 is rolling out on the Chrome Web Store, and Chrome installs it on its own. It carries the extension side of this release: platform guidance from the first step, the new widget and dialog handling, sheet batch mode, next steps after a run, whole reads of background tabs, faster runs from AI apps, and the search and typing fixes. Extensions 38.3.3 to 38.3.7 shipped earlier this week. Nothing in it needs new permissions.

Secure logins. Save a site's username, password and optional 2FA key once, from the Logins page in Cloud or the side panel's Secure logins panel, where the site you are on is one click away. At that site's sign-in page, a Cloud run types the login, makes the 2FA code and keeps the session for later runs; the model never sees the password, and email or SMS codes, captchas and passkeys still come to you. Each site shows a plain status (Synced from Chrome, Kept from cloud sign-in, Signed out by the site), and saving a second login for the same account offers to update the existing one.

WordPress on any site. wp-admin, Elementor, WooCommerce and cPanel get the agent's guidance on every host, where what it learned on one site used to stay there. It builds an Elementor page as real widgets in one go, creates WooCommerce variations in one batch, and fills Rank Math, Yoast and ACF fields without fighting the editor: in our tests a 10-widget Elementor page took 53 seconds and under a credit. For WordPress.com and Jetpack sites, the agent can offer to connect the WordPress.com connector instead. In Cloud and in the extension.

Job portals, course sites, back offices and seller centres on any domain. Guidance now follows the platform, not the address: job applications on Workday, Greenhouse, Lever, SmartRecruiters, Ashby, iCIMS, Taleo, SuccessFactors and Dayforce (including forms embedded on an employer's careers page); Canvas, Brightspace, Blackboard and Moodle on school domains; n8n, GoHighLevel and Odoo on any host; Xero, Amazon Seller Central and Ads, Meta's business tools; and seller centres for Etsy, eBay, Walmart, Meesho, Shopee, TikTok Shop, Flipkart, Lazada and Trendyol. Replaying failed runs, the right guidance was there from the first step in 71 of 84 job and course runs (from 51) and 79 of 90 seller runs (from 37). Your own skills can now target a platform too, such as all Workday job sites. Naming a site in your prompt also brings its guidance from the first step in more languages, including Spanish, Italian, Portuguese, Vietnamese, Indonesian, and Hindi or Bengali typed in Latin letters: 70% of such prompts in our replays, from 47%.

Date pickers, typeahead lists, code editors, grids and search boxes. A date fills a calendar in one call, including month and year menus and both ends of a range; typeahead lists find options that load late; Monaco, CodeMirror, Ace and VS Code for the Web are written through the editor itself, so a long file goes in whole and undoes in one step; and scrolling a long virtual grid reads every row. Calendars now also open in tabs you are not looking at, and a picker or grid that only updates on screen says to bring its tab forward. A dropdown option already chosen is not clicked again in a background tab, so the click no longer lands on the next field. A search box that stands alone is submitted with Enter, and a box that already holds the query moves on to the next step instead of being retyped. In Cloud and in the extension; needs extension 38.3.8 in the extension.

Back-office runs stop tripping over the browser. We went through hundreds of failed runs from people managing stores, sites and admin panels and fixed the causes that were ours. A JavaScript alert, confirm or prompt is answered the way the task says instead of freezing the tab, including one a page shows while it is still loading (on https pages). A tab that stops responding is reopened fresh. An approval you gave covers the run, notices that leave the page usable no longer stop it, and cookie banners are declined. A form missing details is filled as far as it can go and asks you once, every change is read back from the live page before the run says done, and a request lost to a dropped connection is sent again. In Cloud and in the extension.

Next steps after a finished run. A completed answer now shows one-click follow-ups, and at most one offer to run the task again daily, on weekdays or weekly at the time you ran it. Offers appear only on a run you are watching, never in Private Mode or on scheduled runs, and you can snooze or turn off the schedule offer. In Cloud and in the extension.

Sheet batch mode. A task that works through a Google Sheet keeps each row's progress in an "rtrvr status" column, writes in blocks, and picks up where it stopped on the next run. It never rewrites a row marked done, skipped or failed, and a job that produces nothing for its first rows stops instead of stamping the whole sheet. In our test a 21-row job wrote all 21 rows in 37 seconds. Rows that found nothing are listed at the end and can be retried in one go, a list you ask to have in a new sheet is written there first and then worked row by row, and a website is checked for a contact page before a field is left blank. In Cloud and in the extension.

Background tabs read whole. Chrome draws nothing in a tab it is not showing, so feeds like YouTube, Reddit and LinkedIn used to read thin there. In the extension the tab now draws the moment the agent reads it, without coming to the front: first reads found 16 to 24 YouTube videos and 27 to 46 Reddit posts, and runs on Amazon, Indeed and Fiverr finished without pulling the window over your work. LinkedIn profiles read whole: a profile the agent opens in a background tab draws itself within about half a second, so the agent's next look has every section (a run listing a profile's sections answered in full in 24 seconds), and a run through six profiles in one background tab finished in 83 seconds. Pages that are already drawn, like Hacker News, are read at once, and lists that load as you scroll fill in without the tab coming forward: a Luma guest list read all 317 guests. In Cloud a tab is shown just until it draws, and four parallel YouTube reads went from 4 or 5 videos each to 22 to 30.

Research builds a list in one step, and search is faster. On paid plans the planner can hand a lead list or an enrichment job to one research run instead of searching and reading each site itself: on 10 sample tasks it took a median 26 seconds and 2.5 credits, with 90 of 100 rows checking out, where a planned run took about 200 seconds and 13.5 credits. It is billed at cost, uses your free monthly data-tool allowance first, needs Web Search on, and is off in Private Mode. News, site: and local searches answer in a median 0.43 seconds instead of 3, and local results carry phone numbers and websites. Live since October 27.

Annual plans and a free daily schedule. Paying yearly gives two months free: Starter $100, Pro $300, Team $1,000 and Scale $5,000 a year, and plan credits still refill every month. The free plan now includes one daily Cloud schedule and one in Chrome, run on Free Mode. Live since October 27.

Questions don't stall a run. A how-many or which-ones question marks a recommended choice, and the side panel takes it after 2 minutes without an answer and says so. A waiting run badges the extension icon and shows a Chrome notification, and how many is asked up front only before a live action. In Cloud, each step shows the question it asked and your answer, and files you attach while answering stay with the run. Live since October 27.

Better ways in, offered. Instead of clicking through Gmail, GitHub, Notion or Slack page by page, the agent can offer to connect the app. It can also offer to share a Google Sheet with it, ask you for a file, or ask you to sign in first. Saying no snoozes that offer for 14 days. Live since October 27.

Getting started is quicker. On a computer, the first-day welcome in Cloud now leads with the extension and a short walkthrough video, with Start in Cloud still one click away, and after your first clean Cloud run one card suggests running it in your own Chrome next time; on a phone, Cloud leads. Right after install, the page the extension opens signs it in as the account you already use on rtrvr.ai, or offers sign-in right there. In the side panel, the empty message box types out real example tasks, and Tab puts the one shown into the box.

Tailscale never switches IP without asking. If the tunnel cannot connect, the run stops before touching a site and offers Retry Tailscale, Use Retriever proxy for this run (at the normal proxy rate) or Cancel, and a retry keeps your prompt, files and login profile. Schedules retry once and skip that run unless you allowed the proxy on that schedule. Startup gives up in about 10 seconds, and each exit node is its own clear choice in the picker.

Chrome logins hold in Cloud. Cookie Sync now carries a site's partitioned cookies and browser storage from one consistent copy, and a login made in the cloud browser stays until you choose Sync from Chrome. Auto refresh checks social sites every 15 minutes and other sites every 30, a partly restored login carries on with a one-click resync, a login you removed never comes back, and Continue after Take control keeps the session you just fixed. Page reads mask password fields.

Cloud asks you to sign in instead of stopping short. After you accept Sign in now, or when a saved login needs a code only you have, the run pauses for Take control instead of ending as a partial result. With a password already saved the card offers Review login, and with several accounts on one site you pick one and Continue keeps your choice.

Schedules refuse what cannot run, and say why. A save that would fail at its first run, such as a page tool with no page to open, is refused with the reason on the editor tab that needs fixing, and tools a schedule cannot run are greyed out. A schedule whose tool broke later can still be renamed, retimed or paused. The schedule editor is reorganised into Task, Timing and Options, and switching views in Cloud no longer blanks the page. A Weekdays preset joins Daily, and switching a weekdays schedule to Daily now runs it every day instead of Monday to Friday.

Answers that match the ask. A top-10 request with a field list returns ten rows, not one (8 of 8 replays, from 1 of 5). The agent ranks candidates before acting on the best one, a task that says apply or pick any password carries that permission to the step that fills the form, and extraction keeps only each item's own values, so a directory's own email and phone stop filling every company row. A result that prints undefined gets one more look, and one failed search no longer fails the whole step.

Clicks, keys and page reads behave. Reading a page no longer leaves it scrolled to the bottom (on X a single read had dropped the first 9 posts). A loose match never clicks Dismiss inside a card the plan meant to open, Space toggles checkboxes and chips, Enter opens a link in the same tab, a late pop-up gets its own tab, and a drag that changed nothing retries as a mouse drag. LinkedIn member lists read whole, the first read sees past a cookie dialog, and a confirm the agent cancelled is accepted on the retry.

Long runs answer faster and cost less. Each planning pass starts from the run's progress so far instead of recounting finished items, the pass after you answer a question reuses its cached history, and resuming over MCP no longer re-sends the whole run. A lead-list run no longer pauses before free lookups or holds its scoring while slow profile scrapes catch up. Page reads drop repeated icons and duplicate long names and never carry page script text or an icon's own stylesheet, about 26% fewer tokens on the largest pages.

For AI apps and the API: check_results pages a long run's steps (one reply went from 560,095 to 37,799 characters) and a resume keeps the run's output schema; the MCP schedule tool edits, pauses and resumes Cloud schedules with no browser connected; cloud_agent takes continue_from and dataInputs, and treats a reused trajectoryId as already started (409 over REST); and a $V: variable named without a value is reported instead of typed, while extract and crawl now accept variables. A run sent to your browser starts and answers about 3 seconds sooner: it waits for its tabs to load instead of a fixed 2 seconds and closes them after the answer, so outside the model a Hacker News top 10 went from 4.9 to 1.8 seconds.

Smaller fixes. GPT-6.1 Terra takes GPT-6.1 Sol's place in the model menus at half the price, and a saved Sol keeps working. readFile reads every Excel column, row and sheet once per run (48.5 seconds became 0.6). A dropped connection right after sending no longer shows "Couldn't reach" for a run that started, and Cancel no longer leaves a run on Running. A finished run ends with its own closing note instead of the model's last reasoning, and Cloud plays its success sound and confetti only for a clean success, never for a failed, cancelled or partial run. A Free Mode run in Cloud checks the daily allowance at each step, as the extension does. Cloud and scheduled Sheets runs check Drive access with a fresh token before asking you to reconnect, a file the model provider cannot fetch no longer fails the step, element labels never include page script text, and site guidance is refreshed for LinkedIn, Pearson MyLab, Zillow, MercadoLibre and Reddit search.

Tailscale Tunnels are in beta. Install Tailscale on a computer that stays on, make it an exit node, then open Routing in Cloud, choose Add a proxy or Tailscale tunnel, then Tailscale, and press Sign in with Tailscale. The tunnel is saved as Home via Tailscale: pick it for one chat or make it your default, and a schedule made from a run keeps its route. Sites see your connection's IP, and the browser's time zone and language match your exit node.

Safe by design. The cloud browser cannot open your router, printer, NAS or any other device on your network or tailnet; only public sites go through the tunnel. If your exit node is offline, the run stops before it starts and says why, and it never falls back to a data-center route. WebRTC stays inside the tunnel, so it cannot reveal the cloud server's address. Traffic uses your own connection, so it costs no proxy credits, and an OAuth client or auth key you connect with is encrypted and never shown again.

Beta limits. A tunnel you signed in with runs one task at a time; connect with a Tailscale OAuth client for parallel runs. The exit node has to stay on and awake, pages load at your upload speed, and the rtrvr device's key expires after 180 days unless you turn off key expiry for it in Tailscale.

Browser profiles. A profile is a named set of saved logins with its own exit location, such as one per client. A run on a profile reads only that profile's logins and starts signed out everywhere else, and a sign-in made during the run is saved back to it. Default keeps working as before: the logins Cookie Sync uploads plus the sign-ins saved during cloud runs. Switch, create, rename and delete profiles from the Logins chip under the composer, and set your account default there. Free on every plan.

Bring your logins in. Import a Playwright storage state, a cookies.txt file or a Cookie header into Default or into a profile, from the Logins menu or over the API. A profile's cookie values are encrypted with that profile's own key, and the API lists names only: values are never returned and there is no export.

For AI apps and the API: manage profiles at /v1/profiles with the same API key as /agent and /scrape, and pass profile on /agent or on the MCP cloud_agent tool to run on one.

Extension 38.2.3 is rolling out on the Chrome Web Store, and Chrome installs it on its own. It carries the extension side of this release: the new Continue rule, Partial and abandoned run endings, honest step counts, answers on more page-read runs, faster Stop, whole page reads when a script reads one tab several times at once, and a stricter tab filter for AI apps. Nothing in it needs new permissions.

Fewer Continue pauses. A run now pauses for Continue after 4 rounds in a row with no measurable progress, where it used to pause after 2, and the 40-round ceiling stops only a run that has stopped making progress. A run that never makes progress still pauses after its 5 free rounds, as before. Replaying a week of runs with the real pause logic, Continue clicks fell 54% and the share of tasks that needed a click went from 9.2% to 4.2%; after a first Continue, most runs had gone on to finish. Every pause now also records why it fired. In Cloud and in the extension; needs extension 38.2.3 in the extension.

Every run says how it ended. A run that kept what it had and stopped short ends as Partial, and the side panel says Partial result instead of Error. A run whose side panel was closed, whose extension restarted, or that another task replaced is now recorded as abandoned instead of sending no ending at all; on October 23 that was about 1 in 5 extension runs. A Cloud, API or MCP run that stopped short is recorded as partial too, and the schedules panel shows a partial run as its own state, not a failure. Needs extension 38.2.3 in the extension.

Counts you can trust. Every browser step now ends with what its own calls did, such as "refresh_page ran 4 times; refused, not run: 1", whether it finished, stopped short or was stopped by the repeat guard, and the planner counts from that instead of the step's own summary. Asked to reload a page 8 times, a run that split the work across two steps used to report 4; it now reports the 8 that ran. In Cloud and in the extension; needs extension 38.2.3 in the extension.

A run that only read a page ends with an answer on more pages. The check that spots a raw page read now recognises pages whose only element ids are on links, such as an author's page on a quotes site or most article pages, so a run that returns only what it read is held for one review that writes the answer, instead of returning the page. A one-pass Cloud run gets the same review. In Cloud and in the extension; needs extension 38.2.3 in the extension.

Stop takes seconds while a script waits. A run stopped inside rtrvr.waitFor or rtrvr.waitForUrl ends at the next check, where it used to take about 40 seconds, and a stopped Cloud run's card shows its final credits. In Cloud and in the extension; needs extension 38.2.3 in the extension.

A resumed Cloud run shows its whole cost. After you answer a question a run asked, its card, live credits, end line and stored record add the phases before the pause; before, a run that cost 4.08 credits could show 2.31. Billing was always correct.

Files saved to rtrvr Files come back with their link. A Cloud run that saves a file to rtrvr Files gets the file's id and link in its result and gives you that, and it never opens rtrvr.ai's own app in the cloud browser to look for it; one run had waited 5 minutes on a sign-in page there. A plan told to save to rtrvr Files no longer saves to Google Drive.

Page reads hold up. In the extension, a plan script's own page read now waits for an app such as Google Maps to draw, as Cloud's does. Several reads of one tab at the same moment each get the whole page, where most used to come back broken, and a new tab is read once it has left its blank starting page. In Cloud and in the extension; needs extension 38.2.3 in the extension.

A click that does nothing says so. When a click had no effect because the element wants another gesture, the hint now starts with "your click did nothing: this target expects a drag" (or a double click, a choice from a list, a scroll) before the call to make, so the agent stops retrying the click. In Cloud and in the extension; needs extension 38.2.3 in the extension.

Free Mode on shared networks. Accounts signed in with Google no longer count toward or get refused by the per-network account limit, so students and coworkers on one school or office network each keep Free Mode; on the website this needs the browser to pass Free Mode's device check. When the network is the reason for a refusal, the message now says network instead of device. Live since October 23.

AI apps see only the tabs they ask for. Over MCP, get_browser_tabs with a domain filter returns only matching tabs; it used to add your active tab even when it did not match. Needs extension 38.2.3.

For AI apps and the API: a page the cloud browser's network route refuses, or a page that cannot be read, answers 422 instead of a server error; a tool that ran in your browser and failed answers 424 device_tool_failed over the Direct API, while a browser that does not respond keeps its 503; a call refused for a zero balance carries a Retry-After that grows while the key keeps retrying; and a prewarm that gives way to a real run answers 409.

Smaller fixes. A pause at a run's credit limit no longer says a result came back with it when none did, and the tab list you resume with is current. Run records keep every step even when a script fires many calls at once. Background-tab notes reach the planner only when a page really drew less there. Cloud no longer warns about too many listeners on long sessions.