Skyvern

What is Skyvern?

Skyvern is an AI browser automation platform that tries to solve a very old web automation problem, scripts break all the time. Instead of relying on brittle selectors and hand-coded rules for every site, Skyvern uses vision models and language models to look at a page more like a person would, identify what matters, and take the next step. The company was founded in 2023 by Suchintan Singh and Shuchang Zheng, went through Y Combinator, is based in San Francisco, and has raised $2.7 million in seed funding. It also has an open source core, which matters because teams can inspect how it works and self-host if cloud is not an option.

The story behind Skyvern is less about flashy AI demos and more about the economics of browser work. Many operations teams still spend hours inside vendor portals, insurance portals, government sites, and internal web apps because APIs either do not exist or do not cover the workflow. Traditional RPA tools can automate some of this, but they often come with high license costs and constant maintenance whenever a site changes. Skyvern was built around a different idea, if the system can understand the page visually and reason through the task, it should survive layout changes better and require less upkeep.

Who uses it? From our research, the strongest fit is teams doing repetitive browser work with real business consequences: invoice collection, procurement, job applications, claims workflows, healthcare admin tasks, permit filings, and data extraction from sites that keep changing. It is also appealing to developers because it offers Python and TypeScript SDKs, a REST API, Playwright-style workflows, and MCP support for AI agents that need to actually use the web.

Key Features

• Vision-based browser control: Skyvern reads websites through screenshots and page context instead of depending only on CSS selectors or XPath. That matters because many web automations fail the moment a frontend team renames a class or moves a button, and Skyvern’s whole pitch is lower maintenance when layouts shift.

• Planner, actor, validator architecture: Skyvern breaks tasks into sub-goals, executes actions, then checks whether the page changed in the expected way before continuing. In practice, this is why it can handle multi-step flows better than simpler “click and hope” agents, and it helped push Skyvern 2.0 to 85.85% on the WebVoyager benchmark.

• Cross-site workflow automation: A single workflow can be used across many portals that all look different, such as supplier invoice pages or government forms. This matters for operations teams because the alternative is often dozens of separate automations, each with its own maintenance burden.

• Authentication handling: Skyvern supports login flows, session persistence, OAuth, CAPTCHA solving, and some 2FA flows including TOTP. For many real business workflows, login is the hard part, not the form itself, so this is one of the features that turns a demo into something teams can run in production.

• Data extraction and file downloads: It can pull structured data from websites and download documents like invoices, statements, or reports. Teams can export results as JSON or CSV, which is important if the browser task is only one step in a larger finance or ops workflow.

• Workflow builder with 17 block types: Beyond prompts, Skyvern offers a visual workflow builder with blocks for navigation, extraction, login, loops, code, and conditional logic. That gives non-developers a way to build repeatable automations while still giving technical teams enough control for more complex flows.

• Python, TypeScript, and REST API support: Developers can integrate Skyvern into existing apps or internal tools instead of forcing users into a separate dashboard. This matters if browser automation is part of a product or an internal operations system, not just a standalone task.

• Open source and self-hosting: Skyvern’s core is open source under AGPL-3.0, and teams can self-host with Docker if they need more control. For security-conscious buyers, this is often the difference between “interesting tool” and “real option.”

• MCP support for AI agents: Skyvern can be used as a browser tool inside agent workflows via Model Context Protocol. This is useful for teams building agents that need to research, submit forms, or retrieve documents on the web, not just answer questions.

• Observability and replay tools: Users can inspect video recordings, screenshots, visible element trees, and run history when a task fails. Browser automation always needs debugging, and these features reduce the mystery when an agent gets stuck halfway through a process.

Use Cases

One of the clearest Skyvern stories is invoice automation. Many finance teams still log into dozens of supplier portals every month, click through inconsistent navigation, filter by date, and download invoices one by one. Skyvern was built for exactly this kind of work. Instead of maintaining a separate script for every vendor, teams can define a workflow like “download March invoices” and let the system log in, navigate, filter, and save the files. In our research, this was repeatedly presented as a core use case because it combines all the painful parts of browser work, authentication, changing layouts, and repetitive clicks.

Healthcare is another strong fit, especially for admin-heavy browser tasks that happen across fragmented portals. Skyvern markets HIPAA-ready enterprise deployment and has been used for things like eligibility verification, claims-related workflows, provider credentialing, and extracting data from EMR-adjacent systems. The value here is not just speed. It is reducing the time staff spend bouncing between websites that were never designed to work together.

Government and compliance workflows also show where visual automation matters. Permit applications, tax submissions, license renewals, and other public-sector forms often involve long multi-page flows and inconsistent site design across jurisdictions. Traditional automation often turns into a maintenance nightmare here. Skyvern’s pitch is that it can reason through those flows even when the exact page structure changes, which is why teams use it for filing tasks that do not have APIs and cannot justify custom engineering for every portal.

There is also a smaller but very visible developer and power-user story around job applications and browser agents. Skyvern has been used to automate job applications by taking a resume and a job URL, then filling in the repetitive fields and handling the application flow. That use case gets attention because it is easy to understand, but it also shows the broader pattern, Skyvern is most useful when a task lives in the browser, has many steps, and is too annoying or unstable for normal scripts.

Strengths and Weaknesses

Strengths:

Skyvern’s biggest strength is that it attacks the maintenance problem directly. Traditional tools like Selenium or Playwright give developers a lot of control, but every site change becomes your problem. Skyvern’s visual reasoning approach is meant to absorb some of that volatility. For teams dealing with vendor portals or public websites that change often, that can be more important than raw speed.

It also performs unusually well on the kinds of tasks businesses actually care about, form filling, clicking through workflows, logging in, and downloading files. On WebVoyager, Skyvern 2.0 reached 85.85%, which is a strong public result. More importantly, the company published detailed evals rather than vague claims, which we think matters when many AI agent benchmarks are hard to verify.

Another strength is flexibility in how teams adopt it. A developer can use the SDK or API. An ops team can use the workflow builder. A security-focused enterprise can self-host. That range is useful because browser automation usually starts as one team’s problem and then spreads across the company.

Open source is also a real advantage here. Buyers can inspect the project, test it deeply, and avoid betting everything on a black-box vendor. Compared with many AI automation startups that only offer a hosted product and polished demos, Skyvern feels more transparent.

Weaknesses:

Skyvern is still browser automation, which means it inherits some of the messiness of the web. Even with a smarter architecture, edge cases exist. Complex anti-bot systems, unusual auth flows, or pages with poor visual clarity can still cause failures. This is not magic, and teams should expect testing and iteration.

It is also browser-only. If your workflow involves SAP GUI, Excel macros, desktop apps, or legacy enterprise software outside the browser, Skyvern is not the whole answer. In those environments, traditional RPA vendors or hybrid automation tools may still be the better fit.

For pure data extraction, Skyvern may be more tool than some teams need. If the task is just scraping stable pages or pulling data from sites with accessible APIs, a simpler scraper or a standard browser script could be cheaper and faster. Skyvern earns its keep when the workflow includes interaction, not just reading.

Finally, the product is newer than incumbents like UiPath. That means a smaller ecosystem, fewer implementation partners, and less long-term proof in heavily regulated enterprise environments. Some buyers will see that as healthy startup velocity. Others will see it as risk.

Pricing

• Free: $0 Includes 1,000 credits per month, API access, AI browser automation, and CAPTCHA solving. This is enough for testing and small experiments, and especially it does not require a credit card.

• Hobby: $29/month Includes 30,000 credits, faster execution, priority support, and webhooks. This tier looks aimed at individual builders or small teams running moderate workflows.

• Pro: $149/month Includes 150,000 credits, team workspaces, advanced workflows, 2FA credential support, and dedicated support. For a startup or ops team running browser tasks regularly, this is likely the practical starting point.

• Enterprise: Custom Includes unlimited credits, self-hosting, HIPAA support, SOC 2 Type II, SSO, SLAs, and an account manager. This is where regulated industries and larger operations teams land.

The key thing to understand is that Skyvern uses credits, not just seat pricing. In the company’s own framing, a typical automation step may cost around $0.05 to $0.10 depending on complexity, and more involved workflows consume more credits. That is very different from traditional RPA pricing, where you often pay large annual bot licenses whether usage is light or heavy.

Compared with incumbents like UiPath, the headline prices look dramatically lower, but visitors should still model their real workload. If you are running thousands of long, multi-step automations with logins, downloads, and retries, usage can add up. The upside is transparency. The hidden cost to watch is not licensing, it is prompt tuning, workflow design, and testing for edge cases. Even with AI automation, someone still owns reliability.

Alternatives

UiPath UiPath is the obvious comparison because it is one of the best-known RPA vendors. It serves enterprises that want broad process automation across desktop apps, enterprise software, and browsers, with a mature governance and services ecosystem. If you need one platform for SAP, Excel, Windows apps, and browser tasks together, UiPath is still hard to ignore. If your problem is specifically unstable browser workflows and high maintenance on web scripts, Skyvern is the more interesting bet.

Automation Anywhere / Blue Prism These tools live in the same traditional RPA world as UiPath. They tend to appeal to larger enterprises with established automation programs, compliance requirements, and internal teams already trained on older automation methods. They bring process maturity and enterprise procurement comfort. Skyvern is the better fit when teams want a lighter, more developer-friendly tool focused on the browser rather than a full RPA stack.

Selenium Selenium is still everywhere because it is free, flexible, and familiar. For developers who want full control and are willing to maintain selectors, it remains useful. But that maintenance burden is exactly why Skyvern exists. If your workflow touches many changing sites, Selenium can become a long-term tax.

Playwright Playwright is a stronger modern baseline than Selenium for many teams, especially for testing and scripted browser control. It is fast and well-designed, and many developers prefer it. Skyvern does not replace Playwright in every case. In fact, it works well for teams that like Playwright but want AI help for sites where strict selectors become fragile.

Browse AI Browse AI is a friendlier option for people who mainly want no-code web scraping and simple browser tasks. It is easier to approach for non-technical users, but it is more limited when workflows become interactive and multi-step. If your main need is extracting data from a page, Browse AI may be enough. If you need logins, downloads, form submissions, and cross-site workflows, Skyvern has a deeper story.

Browserbase Browserbase is infrastructure, not really a direct substitute. It gives teams managed browsers, session handling, and the plumbing needed to run browser automation reliably. Developers who want to build their own agents may choose Browserbase plus custom logic. Skyvern is what you pick when you want the intelligence layer as well, not just the browser runtime.

Claude Computer Use and similar agent tools General-purpose computer-use agents are getting closer to Skyvern’s territory. They can click around a UI and reason through tasks. But they are broad tools, not specialized browser automation products. Skyvern feels more focused on production browser workflows, especially write-heavy tasks like forms and downloads. A general agent may be better for experimentation. Skyvern is better when the browser task itself is the product.

FAQ

What is Skyvern used for?

Skyvern is used to automate browser tasks like filling forms, downloading invoices, extracting data, logging into portals, and completing multi-step workflows on websites that may not have APIs.

How is Skyvern different from Selenium or Playwright?

Selenium and Playwright usually rely on selectors and custom scripts for each site. Skyvern uses AI vision and reasoning to understand pages more like a human, which can reduce maintenance when websites change.

Is Skyvern open source?

Yes. Skyvern has an open source core under the AGPL-3.0 license, and teams can self-host it if they need more control over data and infrastructure.

Can Skyvern handle logins and 2FA?

Yes, it supports authentication flows including stored credentials, session persistence, CAPTCHA solving, and some 2FA methods like TOTP.

Does Skyvern work for data extraction only?

It can, but that is not always the best reason to choose it. Skyvern is strongest when extraction is part of a larger interactive workflow, such as logging in, filtering, downloading, and exporting results.

How do I get started?

The fastest path is the hosted product. Create an account, get an API key, and test a simple workflow through the app, SDK, or API. If your team needs tighter control, you can evaluate the open source self-hosted route.

How long does it take to set up?

A basic cloud setup can take a few minutes. A real production workflow usually takes longer because you will need to test edge cases, credentials, completion conditions, and failure handling.

Can non-developers use Skyvern?

Yes, to a point. The workflow builder helps non-technical teams create automations, but more complex workflows still benefit from someone technical who can debug and refine them.

Is Skyvern good for enterprise use?

It can be. Enterprise plans include self-hosting, SSO, SOC 2 Type II, HIPAA support, and SLAs. The bigger question is whether your workflow is mostly browser-based, because that is where Skyvern is strongest.

Does Skyvern work outside the browser?

No, not as a general desktop automation tool. If your process depends on desktop software or legacy non-browser apps, you will likely need another tool alongside it.

How reliable is Skyvern?

It is more adaptable than selector-based automation on changing websites, and its 85.85% WebVoyager result is strong. But reliability still depends on the site, the workflow, and how much testing you do before rolling it into production.

Who should consider an alternative?

If you need full desktop RPA, already have a mature UiPath-style program, or only need simple scraping from stable websites, another tool may fit better. Skyvern stands out when the hard part is interacting with messy, changing websites.