The scientific research process is still stuck in the "multi-tool patchwork" era: ClawsGO Science encapsulates retrieval, modeling, drawing and writing into a complete cloud-based scientific research machine.
ClawsGO Science reconstructs scientific research workflows with cloud-native intelligent agents, while Bit Miracle Technology is seeking the next round of financing
The output forms of scientific research work have long been highly digitalized, yet the process itself still remains in the era of assembling multiple discrete tools. A single research topic usually covers multiple links including literature retrieval and reading, data cleaning and statistical modeling, chart drawing, paper writing and typesetting, submission format adjustment and revision. Each link has mature independent tools, but there is no connected execution path. The time researchers spend on environment configuration, data migration, format alignment and version backtracking is often equivalent to the time they invest in creative work, yet these links do not constitute the core value of academic achievements.
ClawsGO Science, a product under Beijing Bit Miracle Technology Co., Ltd., aims to streamline this full path. The product integrates cloud servers, pre-installed scientific research environments, literature retrieval, code execution, data visualization and LaTeX workspace into a single agent interface. Users only need to describe their research goals in one sentence, and the agent will autonomously break down the goals and push forward the whole process continuously. Users can go offline in the middle, and come back several hours later to review the phased results with full process traces. ClawsGO Science is currently in the early commercialization stage after its public launch, and the operating entity is advancing the next round of financing. The raised funds are planned to be used for optimizing model scheduling capabilities, promoting enterprise private deployment and expanding the scientific research team.
Scientific research workflows are fragmented by tool chains, and long-horizon tasks lack a stable execution environment
The pain point in scientific research scenarios is not the lack of single-function tools, but the absence of context sharing between different tools. The literature abstracts retrieved by researchers, cleaned experimental data, modeling scripts and chart files are often scattered in local directories, browser bookmarks and collaboration platforms. When switching devices or resuming work the next day, researchers have to restore the environment and variable status first. When the task chain is extended, once an error or dependency missing occurs in the middle, researchers have to troubleshoot from the very beginning, which not only consumes their attention, but also makes the process hard to reuse.
Such problems have become more prominent after the popularization of large language model programming assistants. General-purpose programming agents are already sufficiently mature in single-step code generation, but their operation cycles are usually designed for one single interaction. When facing scientific research tasks, they are prone to memory loss or interruption in multiple jumps between retrieval, reading, modeling, drawing and writing. Scientific research tasks also have two additional requirements: first, the conclusions and data must be traceable to their sources; second, the operating environment must be fully reproducible. Simply moving the code generation capability into the browser cannot solve the long-horizon nature and evidence chain requirements of scientific research work.
Policy support for scientific research efficiency tools has increased significantly recently. In June 2026, the National Data Administration issued the Implementation Plan for Promoting High-Quality Industry Dataset Construction Action, which for the first time deployed high-quality data construction in fields including scientific research at the industry dataset level, and mentioned making layout around cutting-edge directions. Meanwhile, market research institution QY Research estimates that the global AI for Science market size reached about 4.538 billion U.S. dollars in 2025, and is expected to reach 26.23 billion U.S. dollars in 2032, with a compound annual growth rate of around 28.9%. Among all product forms, software platforms are one of the most important categories. Demands, policy support and the decline of computing power costs are converging in the same time window, making scientific research agents an independent product layer beyond the capabilities of foundational models.
Integrated execution from retrieval to paper completion, the agent runs continuously on the cloud
The product form of ClawsGO Science can be summarized in one sentence: it encapsulates the environments, tools and workflows that originally required researchers to manually connect locally, into a complete scientific research machine running on the cloud. The product allocates a cloud server pre-installed with a scientific research environment for each task, so users do not need to configure LaTeX distributions, Python dependencies or drawing libraries by themselves. Users submit tasks in natural language, such as conducting a certain literature review, modeling a batch of experimental data, or drafting a certain chapter of a paper. The agent then autonomously completes retrieval, reading, code execution and document output, with the whole process running on the cloud. The task status can be viewed synchronously on web pages, mobile terminals and desktop terminals.
At the architecture level, ClawsGO Science adds two layers of engineering capabilities on the basis of the general agent kernel. The first is the long-horizon execution mechanism: tasks can run continuously for several hours around the research goal, and resume from the breakpoint after interruption, without re-executing the previous steps. The second is the task graph and multi-sub-agent collaboration mechanism: complex tasks are split into sub-tasks that can be advanced in parallel, and users can view the overall progress and the execution status of each sub-task on the panel. This design targets the characteristics of long scientific research task chains and easy failure of single steps, which is also the underlying logical difference between it and general programming assistants.
In terms of model scheduling, ClawsGO Science supports switching between multiple cutting-edge models according to task requirements, instead of binding the product to a single model. This design reduces the migration cost caused by model iteration, and also means that the upper limit of product capabilities will be improved synchronously with the foundational models. In the writing session, the product provides a native LaTeX workspace, where the source code and PDF are compiled in real time for comparison. The agent can directly modify the paper: replacing journal templates, adjusting reference formats, supplementing methods and experiments according to review comments can all be triggered by natural language instructions. Modifications of scientific research drawings also support selection-based interaction, and the corresponding code, data and operating environment records will be retained after changes.
Traceable process and reproducible versions are two key capabilities of ClawsGO Science for scientific research scenarios. The environment, commands, file modifications and retrieval actions of each execution will be retained, so researchers can check back at any time, and retrieve historical versions to continue modifications in the revision stage several weeks later. Key conclusions and data are labeled with supporting sources, making the results easy to verify and review. In the operation demo displayed on the official website of the product, the agent can autonomously complete literature retrieval, full-text reading, effect size extraction, random effect model fitting, forest plot drawing and result chapter writing, which takes about dozens of minutes. These are product demonstration examples, not equivalent to the effect commitment for all disciplines and research scales.
Penetrating into scientific research teams with subscription models, evolving from individual efficiency improvement to institutional deployment
The target users of ClawsGO Science cover graduate students, university teachers and independent researchers, as well as enterprise R&D teams. The main demand of the former group is to shorten the daily scientific research workflow and leave more time for topic selection, judgment and direction decision-making; the latter group pays more attention to process standardization, data isolation and private deployment. The two types of demands are connected in product form, but need to be delivered separately: individual scenarios are dominated by self-service subscriptions, while institutional scenarios require pre-integration and continuous operation and maintenance.
In terms of business model, the product takes cloud agent invocation and computing power services as the main charging units. Users pay within their own accounts according to usage volume or packages, the server running duration is not charged separately, and the computing power cost is integrated into the model invocation service. This billing structure limits the variable cost of researchers to the task itself, and does not require them to maintain idle computing power for a long time, avoiding the one-time investment of local high-performance devices. For institutional clients, the team plans to provide private deployment and pre-deployment engineer co-construction mode, which precipitates high-consumption workflows into reusable private assets, and obtains higher unit customer value and renewal stickiness through strong delivery capabilities.
At the present stage, ClawsGO Science is still in the parallel stage of product verification and early commercialization. The product has been publicly launched and accepts user registrations. During the internal test period, the team received feedback from scientific researchers and teachers at universities at home and abroad, and continuously adjusted the efficiency and reliability of tasks accordingly. The specific scale of paying users, repurchase rate and revenue data have not been publicly disclosed yet. The operating entity, Beijing Bit Miracle Technology Co., Ltd., was founded in November 2024, with a registered capital of 100,000 RMB. Its legal representative is Lü Haoran, and its industrial and commercial registration business scope covers software development and artificial intelligence application software development. The team also has a subsidiary Bit Miracle (Changsha) Technology Co., Ltd., which was listed on the Hunan Equity Exchange College Student Innovation and Entrepreneurship Special Board in August 2026, with the listing code 310293HN. Its another product line InfStudio focuses on AI video and clip creation, which together with ClawsGO forms the company's product matrix for content creation and enterprise intelligent scenarios.
The funds raised in the next round of financing are mainly planned to be used for three aspects: engineering optimization of long-horizon task scheduling and multi-model routing, productization of enterprise-level private deployment, and expansion of the scientific research team. Competition in this track is extending from general programming assistants to specialized scenarios, with both foundational model vendors expanding into the scientific research field and vertical startups cutting in from experimental design and literature mining. The unique position of ClawsGO Science is that it takes the full scientific research workflow as the product boundary, instead of only providing a single point function. Whether it can build a good reputation in data compliance, deployment efficiency and process reusability on the institutional side will determine its speed of upgrading from an individual efficiency tool to scientific research infrastructure.
For now, ClawsGO Science still needs to answer several questions with public indicators: the success rate and interruption recovery performance of long-horizon tasks, the versatility across different disciplinary scenarios, and the retention data of paid teams. For products like scientific research agents, how much workflow researchers are willing to hand over to agents, and which links they are willing to pay for continuously, are closer to the quality of commercialization than the number of registered users. This is also the part that Bit Miracle Technology needs to provide verification to the market in the next financing cycle.