Simular is redefining the landscape of digital automation and human-computer interaction by building open, modular, and adaptive AI agents that use computers as intuitively as a human would. At its core, Simular’s technology enables artificial intelligence to perceive, reason, and act within graphical user interfaces (GUIs) across desktops, browsers, and mobile devices, executing complex multi-step tasks with the same flexibility and adaptability as a skilled human assistant. This approach moves beyond traditional API-based automation by empowering agents to interact directly with software through mouse and keyboard control, making them universally applicable even in environments where APIs do not exist or cannot keep pace with the rapid proliferation of new applications.

The foundation of Simular’s offering is its agentic framework, exemplified by Agent S and its successor Agent S2. These frameworks integrate experience-augmented hierarchical planning, continual memory mechanisms, and modular orchestration of specialized and generalist models. By learning from both external web knowledge and accumulated user experiences, Simular’s agents break down complex objectives into manageable subtasks, dynamically adapt to real-time changes, and improve their efficiency over time. The architecture’s modularity allows for seamless integration of new expert modules or foundation models, supporting rapid adaptation to new domains and workflows without the rigidity of monolithic systems.

A distinguishing feature of Simular’s technology lies in its proactive planning and visual grounding capabilities. Rather than relying solely on structured accessibility data, Agent S2 operates directly on raw screenshots, leveraging advanced visual models to locate and manipulate UI elements with high precision. This enables fine-grained control and robust generalization across diverse operating systems and software environments. Furthermore, Simular’s agent-computer interface offloads low-level execution to specialized modules, allowing large language models to focus on strategic planning and decision-making. The continual learning memory system ensures that agents not only recall prior actions but also refine future strategies based on historical successes and failures, creating a foundation for long-term adaptive intelligence.

Simular’s practical impact is evident in its state-of-the-art performance on industry benchmarks such as OSWorld and AndroidWorld, where it surpasses leading alternatives like OpenAI’s CUA/Operator and ByteDance’s UI-TARS in both accuracy and scalability. The framework’s open-source nature fosters transparency, extensibility, and community-driven innovation, setting it apart from closed proprietary solutions. Its cross-platform compatibility extends from macOS-native agents to Windows and Linux environments, while its flexible deployment options—ranging from local execution for privacy-sensitive workflows to cloud-based scalability—address a broad spectrum of user needs.

What differentiates Simular most sharply from competitors is its holistic vision: rather than building yet another digital assistant or task-specific automation tool, Simular aims to create the brain of autonomous computers. By mirroring the modularity and adaptability of the human mind, Simular’s agents are designed not only to automate repetitive tasks but also to collaborate with users in real time, learn from ongoing interactions, and handle both routine and novel challenges with increasing autonomy. This paradigm shift enables organizations and individuals alike to reclaim time spent on mundane digital work, unlock new productivity frontiers, and trust AI systems to operate safely and transparently within their digital environments.

Simular delivers a suite of AI-powered products that enable computers to be used as intuitively as a human would, with a core focus on open, modular, and adaptive agentic automation. At the heart of the offering is Agent S2, an advanced agentic framework that empowers artificial intelligence to perceive, reason, and act within graphical user interfaces across desktops, browsers, and mobile devices. This framework allows AI agents to execute complex, multi-step digital tasks by directly interacting with software through mouse and keyboard controls, bypassing the need for traditional APIs.

The modular architecture of Agent S2 supports seamless integration of both specialized and generalist models, enabling rapid adaptation to new workflows and domains while supporting continual learning from user experience and external knowledge.

The product portfolio includes several distinct applications tailored to different user needs. Simular for macOS provides a native AI agent for Mac users, streamlining daily workflows by automating repetitive actions, managing files, integrating with macOS shortcuts, and supporting natural language commands. This application emphasizes privacy by supporting local execution and offers customizable plugins for deeper integration with popular Mac applications.

Simular Desktop extends these capabilities across platforms, offering a personal AI assistant that records, shares, and replays digital actions as text-based instructions. It is designed for both individual and team productivity, supporting collaborative workflows and offering flexible membership plans that range from free community access to premium private and server-hosted options.

For organizations seeking to automate business operations at scale, Simular for Business introduces an AI workplace assistant that optimizes digital workflows across diverse industries. This solution automates routine tasks, enhances team collaboration, and integrates with existing business systems through transparent scripting and workflow management tools. The platform’s cloud-based deployment options allow businesses to access AI-driven automation from anywhere while maintaining control over data and processes.

Simular also offers the Simular Browser, an AI-powered agentic browser capable of navigating the web, performing research automation, data gathering, and intelligent web navigation through natural language queries. Multi-tab parallel browsing and automated form filling are among its key features, making it suitable for both personal productivity and business research scenarios.

Across all products, users benefit from proactive planning, visual grounding for precise UI interaction, and a continual memory system that enables the agents to learn from past actions and adapt strategies for future tasks. The open-source nature of the core frameworks fosters extensibility and community-driven innovation, while cross-platform compatibility ensures accessibility for macOS, Windows, and Linux users alike.

Flexible pricing models—including free trials, subscriptions, and enterprise add-ons—make Simular’s advanced AI automation accessible to individuals, teams, and organizations seeking to reclaim time from digital drudgery and unlock new productivity frontiers.

Simular’s technical foundation is built on a modular, open-source agentic framework that orchestrates advanced AI agents capable of perceiving, reasoning, and acting within graphical user interfaces across operating systems and device types. At the core, the Agent S2 framework leverages a layered architecture that separates high-level strategic planning from low-level execution, enabling agents to interact with software environments through direct manipulation of mouse and keyboard events rather than relying on brittle API integrations. This approach ensures broad applicability and robust generalization, even in rapidly evolving or API-less software ecosystems.

The back end integrates large language models (LLMs) such as GPT-4o, Claude 3.5 Sonnet, and Deepseek, which are responsible for high-level reasoning, task decomposition, and natural language understanding. These LLMs interface with specialized expert modules—dedicated subsystems optimized for precise visual grounding, UI element recognition, and granular control actions. Notably, Agent S2 operates directly on raw screenshots as its primary input for UI comprehension, bypassing the need for structured accessibility data or DOM trees. Visual models process these screenshots to identify actionable elements, while the agent-computer interface (ACI) translates high-level intentions into executable sequences of low-level actions.

Experience-augmented hierarchical planning distinguishes Simular’s architecture from conventional agentic systems. The planning module draws on both external web knowledge and internal memory systems—specifically, narrative memory for high-level experience recall and episodic memory for detailed step-by-step guidance. This continual memory mechanism allows agents to learn from both successful and failed attempts, refining their strategies over time and supporting long-horizon workflows. The self-supervised exploration phase initializes the agent’s memory with a diverse set of experiences, while ongoing continual updates ensure adaptation to new tasks and environments.

To facilitate extensibility and rapid domain adaptation, Simular’s modular design allows seamless integration of new foundation models or expert modules without architectural overhaul. The orchestration layer dynamically assigns subtasks to the most suitable module or model based on performance characteristics and domain requirements. This design not only optimizes accuracy but also enables the system to scale efficiently as new capabilities are added.

Cross-platform compatibility is achieved through native support for macOS, Windows, and Linux environments, with deployment options ranging from local execution—prioritizing privacy and data residency—to scalable cloud-based operation for enterprise workloads. The Simular Browser extends these capabilities to web automation, embedding the agentic framework within a browser context to enable intelligent navigation, research automation, and multi-tab workflows driven by natural language commands.

From an infrastructure perspective, Simular leverages modern cloud platforms such as Google Cloud for scalable compute resources and distributed training of its visual and language models. The open-source codebase is maintained on GitHub, fostering community-driven development and transparency in both research and implementation.

Looking ahead, the technical roadmap emphasizes further modularization of the agentic stack, deeper integration of proactive planning mechanisms, expansion of the plugin ecosystem for broader application support (especially on macOS), and enhancements to the continual learning system for more personalized automation. Planned upgrades include improved voice command capabilities, expanded support for collaborative cloud-based workflows, and ongoing optimization of the visual grounding pipeline to handle increasingly complex UI scenarios. These developments aim to solidify Simular’s position as a leading platform for adaptive, human-like digital automation.

Based on a review of the initial TAM analyses, the most credible and consistent estimates for the 2024 total addressable market for agentic computer-use automation platforms, including Simular, fall within the range of $4.2 billion to $7.2 billion. This range is derived by excluding outlier analyses that are significantly lower ($2.4b-$3.1b) or higher ($7.2b-$10.6b) than the consensus of the majority of sources. The remaining analyses cluster tightly between $4.2 billion and $7.2 billion, with multiple independent estimates supporting this interval. Top-down analysis begins with the global market for intelligent automation and digital workforce solutions, which is estimated to exceed $20 billion in 2024 according to IDC, Gartner, and MarketsandMarkets, but only a subset of this market—focused on open, modular agentic frameworks capable of direct GUI interaction across consumer and enterprise endpoints—applies to Simular’s core offering. Within this segment, direct competitors such as OpenAI CUA/Operator, ByteDance UI-TARS, and Microsoft Copilot target a combined market that several industry reports (e.g., Grand View Research, Everest Group) size at approximately $5 billion to $7 billion in 2024 for advanced agentic desktop and browser automation. This is corroborated by the funding levels and revenue multiples observed for leading players in the sector, as well as by adoption rates among enterprise and professional users. Bottoms-up analysis considers the number of potential addressable users across macOS, Windows, and Linux platforms. There are over 1.5 billion desktop OS users globally (StatCounter), with an estimated 300 million knowledge workers (World Bank, OECD) who regularly engage in digital workflows that can benefit from agentic automation. Assuming a conservative penetration rate of 2% to 3% for advanced agentic solutions in 2024, and an average annual spend per user (across individual, team, and enterprise tiers) of $100 to $150 (based on published pricing from Simular, OpenAI Operator, Copilot, and Manus AI), the bottoms-up TAM calculation yields a range of $4.5 billion to $6.75 billion. This aligns closely with the top-down findings and further supports the $4.2 billion to $7.2 billion range as a robust estimate for Simular’s 2024 TAM. Data sources include StatCounter GlobalStats for OS user base estimates, World Bank and OECD for knowledge worker counts, pricing data from Simular’s public site and competitors’ disclosures, and market sizing from IDC, Gartner, Everest Group, and Grand View Research. The final TAM range reflects a blended approach that is consistent with both industry standards and observed market dynamics.

In a landscape crowded with agentic automation platforms, Simular distinguishes itself through a combination of open-source modularity, experience-augmented hierarchical planning, and a uniquely human-centric approach to digital task automation. Unlike OpenAI CUA/Operator and ByteDance UI-TARS, which are both closed-source and tightly coupled to their respective ecosystems, Simular’s Agent S2 framework is fully open and community-driven, enabling rapid extensibility and transparent innovation.

This open architecture not only fosters trust and adaptability but also allows seamless integration of new expert modules and foundation models, making it possible for users and developers to tailor the system to emerging workflows or industry-specific requirements without vendor lock-in or architectural rigidity.

Whereas competitors such as OpenAI CUA/Operator and Microsoft Copilot rely heavily on monolithic LLM-driven reasoning and API integrations, Simular’s agentic stack leverages a layered design that separates high-level strategic planning from low-level execution. The experience-augmented hierarchical planning module draws on both external web knowledge and internal continual memory—comprising narrative and episodic memory systems—to enable agents to learn from both successful and failed attempts, adapt to novel domains, and refine strategies over time.

This continual learning mechanism is notably absent in ByteDance UI-TARS, Manus AI, and Auto-GPT, which typically lack persistent, user-specific memory or proactive adaptation across sessions.

Simular’s approach to visual grounding further sets it apart. While ByteDance UI-TARS and OpenAI CUA/Operator utilize visual input for UI understanding, Simular’s agents operate directly on raw screenshots rather than relying on accessibility trees or structured DOM data. This enables robust generalization across diverse software environments—including those lacking accessibility support or standardized APIs—and supports fine-grained control of UI elements.

The modular orchestration layer dynamically assigns subtasks to the most capable module or model, optimizing for accuracy and efficiency in real time, a flexibility that monolithic competitors cannot match.

Cross-platform compatibility is another area where Simular excels. Native support for macOS, Windows, and Linux—alongside flexible deployment options ranging from local privacy-preserving execution to scalable cloud-based operation—contrasts with the more siloed offerings from Microsoft Copilot (Windows-centric) and ByteDance UI-TARS (Android and Windows focus).

The Simular Browser extends these capabilities to web automation with intelligent navigation, research automation, and multi-tab workflows, all driven by natural language commands and benefiting from the same proactive planning and continual learning found in the desktop agents.

From a user empowerment perspective, Simular’s philosophy centers on building adaptive collaborators rather than static workflow automators. Agents learn from user interactions, adapt to evolving digital environments, and proactively suggest optimizations—moving beyond the repetitive task automation focus of Manus AI or the rigid workflow scripting of Auto-GPT.

The product suite’s flexible pricing models, transparent scripting via Simulang code, and active community engagement further lower barriers for individuals, teams, and enterprises seeking to reclaim time from digital drudgery.

Taken together, these differentiators position Simular as the most open, adaptive, and extensible agentic automation platform in its class—delivering state-of-the-art benchmark performance while remaining accessible to a broad spectrum of users and developers.

The leadership and technical team behind Simular demonstrates a strong concentration of academic and industry experience in artificial intelligence, agentic frameworks, and computer-use automation. Ang Li, the founder and CEO, brings an impressive pedigree to the company, having previously served as a research scientist at Google DeepMind and as Head of US Software Engineering at Baidu Apollo. He holds a PhD in Computer Science from the University of Maryland, was a research associate in robotics at Carnegie Mellon University, and earned his undergraduate degree from Nanjing University. Li’s publication record is extensive, with dozens of papers in top AI venues, and he has led teams in both research and engineering capacities at major technology firms. His background in reinforcement learning and large language models underpins Simular’s technical direction.

Jiachen (JC) Yang, co-founder and Chief Technology Officer, is an AI researcher with a focus on agentic frameworks and multi-agent systems. He completed his PhD in Computer Science at the University of California, Berkeley, specializing in reinforcement learning and artificial intelligence. Yang’s prior experience includes roles at Meta AI Research and other prominent technology companies, and he has published widely in leading AI conferences.

Among the founding engineers, Chih-Lun Lee stands out for his work on backend infrastructure and AI model integration. Lee holds a master’s degree in Computer Science from National Taiwan University and has practical experience from several startups. Another founding engineer, Alex Covaci, contributes expertise in cross-platform application development and user interface design, with a degree from the University of Waterloo.

Hao Liu serves as a founding research scientist, focusing on reinforcement learning and agent-based modeling. Liu obtained his PhD in Computer Science from the University of California, Berkeley, with research stints at DeepMind. His academic record is strong, though his recent professional appointments appear to be in academia rather than industry.

The research team includes Saaket Agashe, who holds a PhD in Computer Science from Georgia Tech and is recognized for his work on cognitive architectures for intelligent agents. Jiuzhou Han is another research scientist whose expertise lies in multimodal large language models; he is currently pursuing a PhD at Monash University after earning a master’s degree from the University of Melbourne. Shuyu Gan contributes to research on memory mechanisms for agentic frameworks and holds a PhD from the University of California, Berkeley.

The advisory board features Xin Eric Wang, an assistant professor at the University of California, Santa Cruz. Wang specializes in natural language processing and multimodal AI, providing academic guidance to Simular’s research efforts.

While the core team boasts significant academic credentials—many with doctorates from top-tier institutions such as UC Berkeley, Georgia Tech, and Monash University—and experience at leading technology companies like Google DeepMind, Baidu Apollo, and Meta AI Research, there are some limitations. Several key contributors are early in their careers or still completing doctoral studies, which may impact the depth of operational experience in scaling enterprise products. The team’s collective experience is heavily research-oriented, with less evidence of prior successful commercial product launches or exits. This academic focus could present challenges when transitioning from research prototypes to robust, user-facing software at scale. Nevertheless, the technical depth and publication record across the leadership and research staff position Simular as a project with strong foundational expertise in AI agentic systems.

Simular’s go-to-market approach leverages a phased rollout of its agentic automation products, targeting both individual users and enterprise organizations through a combination of open-source community engagement, direct downloads, and strategic partnerships. The initial user acquisition strategy centers on offering free trials and public betas for core applications such as Simular for macOS and Simular Browser, which are accessible via the company’s website and promoted through channels like Product Hunt, Discord, and social media platforms including LinkedIn and X. Early access programs and waitlists for products like Simular Desktop help build anticipation while capturing leads for future premium offerings. For business customers, Simular positions its AI workplace assistant as a solution for automating digital workflows across industries, emphasizing integration with existing business systems and transparent scripting for workflow management. The company encourages organizations to request demos and join early-access programs, facilitating direct engagement with decision-makers in sectors such as healthcare, finance, insurance, device manufacturing, and recruiting. Marketing efforts include content-driven campaigns through the company blog, technical articles, and comparative analyses that highlight Simular’s differentiation from competitors like OpenAI CUA/Operator, ByteDance UI-TARS, and Microsoft Copilot. Simular also benefits from recognition in industry benchmarks and awards, such as the ICLR Agentic AI workshop Best Paper Award, which bolsters credibility among technical audiences. Community-driven growth is supported by open-source releases of core frameworks like Agent S2 on GitHub, fostering developer adoption and contributions of new expert modules for specialized use cases. The company actively participates in accelerator programs, such as the Google Cloud AI Accelerator, to expand its reach and access technical resources for scalable deployment. Pricing models are designed to lower barriers to entry, with free plans for individual users and tiered subscriptions for advanced features, team collaboration, and server-hosted options. Premium add-ons and enterprise concierge services cater to organizations seeking tailored automation solutions. Simular’s roadmap includes expanding plugin support for macOS applications, enhancing voice command capabilities, and deepening integration with productivity platforms through partnerships—exemplified by its collaboration with Notion. These initiatives aim to drive user engagement, increase market penetration in both consumer and enterprise segments, and establish Simular as the leading open, adaptive agentic automation platform.

Momentum for Simular’s adoption has accelerated through a multi-pronged strategy that combines open-source community engagement, direct user onboarding, and targeted outreach to both individuals and enterprises. Public beta releases of Simular for macOS and Simular Browser have driven early user acquisition, with downloads made easily accessible via the company’s website and promoted through high-visibility channels such as Product Hunt, where Simular achieved the #2 daily rank on April 26, 2025, garnered 393 upvotes, and attracted over 36 user comments within days of launch. The platform’s Discord community has grown to over 600 members, while more than 1,265 users have joined the Simular for Business early-access program, reflecting strong initial traction among both consumers and organizational teams seeking workflow automation. Early access and waitlist programs for Simular Desktop and premium features have further built anticipation and captured leads for future conversion. The phased rollout approach includes free trials and tiered subscription plans, lowering barriers for individuals to experiment with the technology before upgrading to paid offerings. For enterprise adoption, Simular leverages demo requests and direct engagement with decision-makers in industries such as healthcare, finance, insurance, device manufacturing, and recruiting, as evidenced by its positioning as a trusted solution across these sectors. The open-source release of core frameworks like Agent S2 on GitHub has fostered developer adoption and community contributions, with new expert modules emerging for specialized domains. Strategic partnerships, such as the integration with Notion’s productivity platform, have expanded Simular’s reach into established business ecosystems. Recognition through industry benchmarks—such as achieving state-of-the-art results on OSWorld and AndroidWorld benchmarks, as well as winning the Best Paper Award at the ICLR 2025 Agentic AI workshop—has bolstered credibility and visibility among technical audiences. The company’s participation in the Google Cloud AI Accelerator has provided additional access to technical resources and a global network of potential adopters. As a result of these efforts, Simular has established itself as a leading open agentic automation platform with a rapidly expanding user base spanning individual professionals, developer communities, and enterprise teams across North America and Europe.