Vercept - AI that works directly on your computer

What Is Vercept

In 2022, we discovered a pain point that had been largely overlooked: the vast majority of AI tools require users to upload their data to the cloud for processing. For privacy-conscious individuals and businesses, this created a significant barrier to adopting AI assistance. Whether it was sensitive customer data, proprietary business information, or confidential financial records, the thought of sending these materials to external servers gave many users pause.

We believed there had to be a better way.

We set out to build a different kind of AI—one that could run directly on your computer, understand your screen content, learn your workflow patterns, and recognize your operational intentions without ever needing to transmit your data to distant data centers. This vision gave birth to Vercept, an AI assistant designed to operate at the edge, combining the power of local edge computing with selective cloud services when needed.

Our product, known internally as "Vy," was more than just a technical experiment. It earned the trust of paying subscribers who saw the value in having AI assistance that respected their data boundaries. Vy became a reliable colleague for professionals who needed intelligent automation but couldn't compromise on data privacy.

In February 2026, our journey took a significant turn when Anthropic announced the acquisition of Vercept. This wasn't simply a business transaction—it represented a meeting of minds. Anthropic shared our belief that AI should be both powerful and responsible, that safety and capability aren't mutually exclusive, and that building controllable AI systems should be the industry's north star.

While Vercept's services will cease on March 25, 2026, this isn't an ending. The work we've pioneered at Vercept will continue at Anthropic, carrying forward our mission to make AI that truly respects users while delivering meaningful assistance.

---

Vercept's Core Capabilities

The magic behind Vercept lay in its ability to function as a true digital colleague—one that could see what you see, understand what you're trying to accomplish, and take action on your behalf. Let's explore the capabilities that made this possible.

Screen Understanding served as the foundation of the entire experience. Vercept could "see" what appeared on your screen, interpreting visual information just as a human would. This wasn't simple screenshot capture—it involved sophisticated computer vision models trained to understand UI elements, text content, image context, and the spatial relationships between different screen components. When Vercept understood what you were looking at, it could provide relevant assistance without requiring you to manually explain your current context.

Workflow Understanding enabled Vercept to become genuinely useful over time. The AI observed and learned your recurring patterns—how you typically handled emails, processed documents, managed projects, or navigated between applications. Rather than offering generic suggestions, Vercept could identify automation opportunities specific to your work style. One of our users told us that after a few weeks, Vercept began anticipating tasks she repeated daily, saving her significant cognitive load.

Intent Recognition represented the bridge between observing and acting. By analyzing your sequence of operations alongside screen content, Vercept could determine what you were genuinely trying to achieve. This meant moving beyond literal commands to understand underlying goals. When you started drafting what looked like a contract follow-up, Vercept recognized the intent and could prepare relevant templates or pull necessary information from your other applications.

AI Action Execution was where assistance transformed into collaboration. When you granted permission, Vercept could perform actual operations on your behalf—agreeing to contract terms you've reviewed, confirming payments after you've approved transactions, or filling standardized forms based on your instructions. This capability required robust safety mechanisms and clear user consent protocols, which we built into the system from the ground up.

Cross-Application Coordination distinguished Vercept from single-purpose tools. Your work rarely happens in isolation—you move between email, spreadsheets, document editors, project management tools, and countless other applications. Vercept understood these boundaries and could coordinate across them, pulling information from your calendar to draft meeting notes, or transferring data from emails into your CRM.

Local Execution with Privacy Protection was perhaps the most distinctive capability. The core AI engine ran locally on your computer, meaning your sensitive data never needed to travel to external servers for processing. Only when specific capabilities required cloud resources—such as demanding computational tasks—would Vercept connect to cloud services, and even then, you maintained control over what data left your device.

---

Who Uses Vercept

Understanding who benefited most from Vercept helps illustrate its real-world value. While we've always believed AI assistance could help broadly, certain user profiles found particularly strong alignment with what Vercept offered.

Privacy-Focused Enterprise Teams represented a significant portion of our user base. These were organizations that recognized AI's potential to boost productivity but operated under strict data handling requirements. Financial services firms, healthcare providers, and legal practices faced regulatory obligations that made uploading sensitive client data to cloud-based AI services problematic. For these teams, Vercept's local processing capability wasn't just a nice-to-have—it was often a prerequisite for adopting AI tools at all.

Individual Professional Users found Vercept invaluable for automating repetitive desktop tasks. Designers managing assets across multiple applications, developers juggling coding environments with documentation, and researchers organizing literature reviews—all discovered that Vercept could handle the tedious portions of their work that didn't require human creativity or judgment. One user in market research told us Vercept saved her roughly two hours daily by automatically compiling and formatting data from various sources into consistent reports.

Knowledge Workers Across Industries benefited from Vercept's ability to coordinate across applications. Modern professional work rarely stays contained within a single tool. You might be drafting a proposal in a document editor while referencing data from a spreadsheet and tracking deadlines in a project management tool. Vercept understood these connections and could help maintain coherence across your digital workspace.

Privacy-Sensitive Industries had particularly compelling use cases. When handling patient records, client confidential information, or proprietary business data, the assurance that AI would process everything locally provided peace of mind that cloud alternatives simply couldn't match. This wasn't about distrusting cloud providers—it's about defense-in-depth and maintaining clear boundaries around sensitive information.

---

Technical Architecture and Core Features

Vercept's technical foundation reflected our core belief that AI should run where your data lives. Here's how we built a system that could understand your screen, learn your workflows, and assist you—all while keeping your data close.

Edge AI Architecture formed the system's backbone. Rather than routing everything through cloud servers, we designed Vercept to run its core AI models locally on your computer. This approach meant that for typical assistance tasks, your data never left your device. Cloud connections occurred only when you specifically requested capabilities that required additional computational resources, and even then, you maintained transparency and control over what was transmitted.

Hybrid Cloud Service Integration provided the flexibility to balance local privacy with powerful capabilities when needed. We integrated services from Amazon Web Services, Google Cloud, and Together.ai to handle specific functions that benefited from cloud-scale resources. This hybrid model meant users didn't sacrifice capability for privacy—they got both.

Screen Vision Understanding System represented one of our most technically demanding achievements. We developed specialized computer vision models capable of interpreting screen content with human-like comprehension. The system could identify UI elements, understand text content, recognize images, and grasp the spatial layout that humans intuitively understand when looking at a screen. This wasn't screen recording—it was sophisticated visual interpretation that enabled truly contextual assistance.

Intent Reasoning Engine powered the transition from observation to action. Using deep learning approaches, we built systems that could analyze your operation sequences and screen context to determine what you were trying to accomplish. This went beyond keyword matching or command recognition to understand genuine user goals—a critical capability for providing helpful rather than intrusive assistance.

Secure Payment Processing through Stripe ensured that subscription management worked reliably for both individual and enterprise customers. We treated payment infrastructure with the same security emphasis as our core AI systems, understanding that trust in financial handling directly impacted user confidence in the entire product.

API Developer Interfaces opened Vercept's capabilities to third-party integration. Developers could incorporate Vercept's understanding and assistance features into their own applications, extending the benefits of screen-aware AI to specialized workflows and use cases.

---

Vercept's Journey and Investment Background

Every innovative product has a story behind it—one that explains not just what was built, but who built it and why they believed in the mission. Vercept's journey reflects the conviction that privacy-respecting AI wasn't just possible, but necessary.

The Founding Team brought together expertise across AI research, system design, and product development. Led by CEO Kiana Ehsani, with co-founders Luca Weihs and Ross Girshick, the team combined academic rigor with practical startup execution. Each founder brought distinct strengths that covered the technical, operational, and strategic dimensions required to pursue such an ambitious vision.

Incubation Support from AI2 Incubator—the prestigious program run by Allen AI—provided critical early-stage backing. AI2 Incubator doesn't just provide funding; it offers deep technical expertise and a network built around responsible AI development. This alignment between our mission and our incubator's focus reinforced our conviction that we were pursuing something genuinely important.

Investment Backing came from firms that understood both the technical challenges and the market opportunity. Fifty Years, Point Nine, and Madrona joined AI2 Incubator in supporting our vision, alongside several angel investors who brought additional perspective and networks. Each investor brought more than capital—they brought belief in the possibility of AI that respected user privacy while delivering genuine assistance.

The Acquisition by Anthropic in February 2026 marked both an ending and a continuation. Anthropic's acquisition wasn't simply a financial exit—it represented validation of our core belief that AI should be built responsibly. Their public statement captured this sentiment: both organizations shared the conviction that AI should be both useful and controllable, that safety and capability must advance together, and that building AI that genuinely serves users requires putting these principles first.

The work begun at Vercept continues at Anthropic. The team's expertise, the technical foundations built, and the philosophy that guided our development—all of this lives on in the broader Anthropic mission. While the Vercept product as known to users will cease operations, the underlying contribution to responsible AI development persists.

---

Frequently Asked Questions