rtrvr.ai
Browser ExtensionStart in Chrome, on the page you're on.CloudA thousand browsers, on your schedule.RoverThe AI customer engineer for your product.Data & EvalsExpert trajectories for AI labs.
ACCESSAPI + MCPCLI & SDKTemplatesIntegrationsWhatsApp
Use cases
Vibe ScrapingLead EnrichmentWeb MonitoringForm FillingJob ApplicationsSocial MediaAI Web ContextAgentic CheckoutAll use cases
Pricing
BlogLaunches, benchmarks, deep divesDocsExtension, Cloud, API, MCP, CLIModelsWhich model runs your taskCase StudiesReal teams, real runsVideosNew runs every weekChangelogWhat just shippedNewslettersProduct releases and real runs
Docs
Log inBook DemoAdd to Chrome
Log in
Menu
Add to ChromeBook a demo
ProductsBrowser ExtensionCloudRoverData & EvalsExploreUse casesPricingBlogDocs
DocsChrome / Cloud / API / MCP
Quick startAPIMCP

Start

OverviewQuick start

Build

Web agentJevSheets workflowsRecordingsTool callingConnectorsSkillsEnrichment datasets

Run

CLI and SDKAPI overviewAgent APIScrape APIBrowser API and MCP

Automate

ShortcutsTriggersWebhooksSchedules

Trust & help

Cookie syncPermissions and privacyFAQ
DocsJev

Try Jev in rtrvr.ai

Let Jev choose browser actions alongside your usual AI model. Turn it on in Chrome or Cloud and try it on a task you know.

Experimental Optional. Off by default.

Turn on Jev

Open the Chrome extension or rtrvr.ai Cloud. Click the model name beside the prompt box, then turn on Try Jev. Keep your preferred model selected.

Chrome extension

1 Click the model name beside the prompt box.

2 Turn on Try Jev.

Cloud

1 Click the model name beside the prompt box.

2 Turn on Try Jev.

The screenshots show Jev on. You can turn it off in the same menu. It is available in Free Mode and paid plans, with no separate TypeSafe account or API key.

What Jev does

Jev is TypeSafe’s model for structured decisions. It chooses from options supplied by software and returns probabilities. In rtrvr, it helps choose the next browser action.

  1. Your selected model plans the task. It writes code and text, handles open-ended steps, and prepares the work.
  2. rtrvr reads the page and builds the choices. Each choice contains a tool call with its parameters already filled in.
  3. Jev selects a choice or hands the step back. Your selected model handles steps that need its capabilities.
  4. rtrvr runs the action and reads the result. The updated page informs the next step.

Turning Jev on lets it participate where supported. Your selected model keeps its full planning context and handles steps that need generation or cannot be expressed by the supplied choices.

How rtrvr prepares an action

rtrvr builds complete choices from the current page: click an observed element, type an exact supplied value, or select an available option. It supplies element IDs and arguments, maps logical tabs to real browser tabs, and checks that the page observation still matches before returning the choice for execution.

Jev’s Choice operation selects one of those calls or delegates. If the action request is too large, rtrvr can first use one batch of Score requests to select relevant page context, then ask Choice once. Planner prompts skip that scoring step. Explicit typed judgments can also use Jev where supported.

Selecting saved site skills with Jev and reusing earlier page-region scores are follow-up ideas. They are not part of this setup.

Try a familiar task

Start with a task whose result you can check. For example:

Find three reusable water bottles under $30 on this site. Return each name, price, and product link.

Run it with Jev off, then on, using the same selected model and starting page. Check the result, elapsed time, and credits used. Repeat on a few tasks you actually need; one run can be affected by changing pages, cached context, and different choices.

Our first LinkedIn and Amazon comparisons finished sooner and had higher estimated inference costs with Jev. Each configuration ran once. The full experiment explains the results and limits; those runs predate the removal of planner-context scoring and do not establish a fixed saving for the current release or other tasks.

Watch the experiment

IN THIS WALKTHROUGHPlay videoSee how Jev fits your next run.

Jev for AI Agents: I Tested it with rtrvr.ai for browser automation

Jev inside rtrvr: the setup, tasks, results, and tradeoffs.

Watch on YouTube or read the findings and see the recorded data.

Questions and fixes

I can’t find the switch.
Open the model picker beside the prompt box and look for Try Jev, below Free Mode. In the extension, update to version 37.9.0 or later and reopen the side panel.
Should I choose Jev from the model list?
Keep the model you normally use. Try Jev is a separate switch that works alongside that choice. See the model guide if you need help choosing your main model.
Why is it paused in Private Mode?
Jev is not covered by rtrvr’s zero-data-retention promise, so Private Mode skips it. Your saved Jev preference is kept.
Do I need a TypeSafe key, or does it make my run free?
No separate TypeSafe key is needed. Your rtrvr plan’s usage rules still apply. Jev is available in Free Mode within its allowance and on paid plans.
Will every task be faster or cheaper?
That depends on the page, the task, your selected model, and the work needed to prepare Jev’s choices. Compare the finished result as well as time and cost.
PreviousWeb agentNextPermissions and privacy

YOUR NEXT RUN

Try it on a task you know.

Open model settings, turn on Try Jev, and compare the result with your usual run.
Add to ChromeRun in CloudBook a demo

On this page

Turn on JevWhat Jev doesTry a familiar taskWatch the experimentQuestions and fixes
rtrvr.ai

Make every site
work for you.

Launches first, roadmap early, and the occasional trick we only share by email.

Products

Browser ExtensionCloudRoverData & Evals

Use cases

Vibe ScrapingLead EnrichmentForm FillingWeb MonitoringSocial MediaJob ApplicationsData MigrationAI Web ContextAgentic Checkout

Resources

DocsBlogModelsData for AI LabsCase StudiesVideosNewslettersChangelogPricingAppSumoDemoAffiliate

Company

TeamContactGCP PartnerWhat We BelieveSecurityPrivacyTerms

Developers

APIMCPCLI & SDKTemplatesIntegrationsWhatsApp

Compare

ApifyBardeenBrowserbaseBrowser UseClayClaudeCometFirecrawl
Products
Browser ExtensionCloudRoverData & Evals
Use cases
Vibe ScrapingLead EnrichmentForm FillingWeb MonitoringSocial MediaJob ApplicationsData MigrationAI Web ContextAgentic Checkout
Resources
DocsBlogModelsData for AI LabsCase StudiesVideosNewslettersChangelogPricingAppSumoDemoAffiliate
Company
TeamContactGCP PartnerWhat We BelieveSecurityPrivacyTerms
Developers
APIMCPCLI & SDKTemplatesIntegrationsWhatsApp
Compare
ApifyBardeenBrowserbaseBrowser UseClayClaudeCometFirecrawl
BACKED BYNVIDIA InceptionGoogle Cloud for StartupsBright DataNEC XSalesforce LaunchpadElevenLabs GrantsGMI CloudComposioSmallest.ai Grants
DISCOVERYllms.txtllms-full.txtagents.mdDocumentation indexSitemapOpenAPIAI Catalog
© 2026 Retriever AI · rtrvr.ai · Cookie settings
DiscordYouTubeInstagramTikTokLinkedInXGitHub
support@rtrvr.ai