ScienceOS is designed to streamline your workflow. The AI tool can access a database of more than 220 million research papers to generate answers with in-text citations, allows for uploading and interacting with PDFs, and offers a built-in reference manager to chat with your personal database of thousands of full-text papers. Advanced algorithms prioritize reliable sources, while intuitive chat-based interactions make the experience feel like you’re collaborating with a colleague. Behind the scenes, scienceOS employs an LLM-based AI agent framework to handle complex research tasks.

The scienceOS AI research agent

An AI agent is a system designed to autonomously complete tasks. At its core, it consists of a central AI that processes input, uses tools to gather or manipulate information, and generates a response. These tools can be highly sophisticated and may even include other AI models. In the case of LLM-based AI agents, the central AI is a Large Language Model (LLM), which enables the agent to understand human language and produce the desired output.

Schematic of the scienceOS AI research agent. A user interacts with the AI research agent by entering a request or uploading a file (e.g. PDF). Then, scienceOS makes use of one or multiple of its tools to generate relevant output.

The AI research agent scienceOS employs the LLM-based AI agent framework and has access to the following three types of tools:

1. The search tool
   The search tool provides scienceOS with access to a database of more than 220 million abstracts. As a first step, the AI agent translates the input of the user into a database-compatible search query. Then, the search tool retrieves up to 100 works from the 220 million papers and forwards them to a complex ranking algorithm. This algorithm factors in metadata such as publication date and citation count but also contains another AI model specialized in evaluating the content relevance. The goal of the ranking algorithm is to identify the top 10 sources according to rigorous academic standards. Before providing these sources to the AI agent, the search tool makes one final adjustment: it creates a citation network to discover important related papers and add them to the final list of sources. From here, the central AI takes over again: it reads the provided sources and decides how to continue.
2. The PDF tool
   The PDF tool becomes available to the AI research agent, once a PDF file has been uploaded to a chat session. During the upload, the PDF is processed by a machine learning-powered document parser. After the upload, the PDF tool provides scienceOS with the ability to review a detailed AI-generated summary of the document but also search for specific details in the full text. Ultimately, the PDF tool will provide the AI agent with relevant text passages, figures, and tables. During this process, the central AI must process a lot of information – especially if a chat contains multiple PDFs. This is why the PDF tool requires access to our best model and is only available during the free trial period or for Angel users.
3. The library tool
   No centralized literature index can be complete. Internal documents, paywalled papers, grey literature – there will always be something missing for your research. This is why you can also upload up to 4,000 PDFs and references to a free built-in AI reference manager: the scienceOS library. Once your library contains sources, the AI research agent gets access to the library tool. Similarly to the PDF tool, the library tool enables scienceOS to get an overview of the sources in your library but also search for details in uploaded PDFs. If you make use of collections to organize the sources in your library, you can also instruct scienceOS to restrict the library tool to a certain collection. In the end, the library tool will forward relevant readings to the central AI. For answers based on full-text papers, this requires our best model; chatting with up to 4,000 PDFs in your library is only available during the free trial period or for Angel users. However, uploading PDFs to your library and chatting with their respective abstracts is available with the free plan.

Irrespective of the tools the AI research agent makes use of, neither your inputs (e.g. your prompts or uploaded files) nor the AI agent outputs (e.g. text answers or image files) are used to train new scienceOS AI models.

Combining multiple tools in one chat

If you upload a PDF to a chat and add some sources to your library, the scienceOS AI research agent has access to all three types of tools at once. This makes scienceOS more powerful; as an AI agent, the AI can autonomously decide or be instructed to use multiple tools for a given task.

How to combine multiple tools of the AI research agent scienceOS. A brief video showing how to combine the library tool and the search tool of the AI research agent to draft and critically review the outline to an introduction.

For example, you may need to write an introduction section for a grant application. First, you upload a set of full-text papers to your scienceOS library – all the relevant literature you are aware of. Then, you instruct scienceOS to draft an outline for the introduction based on the papers in your library. Finally, you direct the AI research agent to critically review the outline based on papers outside your library. By combining the use of the library tool and the search tool in the same chat, scienceOS helps you to review yourself and answer questions such as: is there something missing, are important contrary views considered, and what are recent trends?

Such robust multi-tool prompt sequences help you become more efficient in your research. Especially if they are supporting frequent tasks in your research workflows. To make continued use of your high value prompts, you may also turn them into scienceOS AI Actions. These customizable saved prompts offer an efficient way to automate your processes. For example, the complex AI Action ‘Critical review’ instructs the AI research agent to use the search tool twice: once for literature that supports the statements made in the previous answer and once for papers that contradict them. With just one click, scienceOS provides a critical assessment of the previous chat message by using multiple tools in one run.

Try scienceOS yourself

The AI research agent is designed to support you with your research tasks. From quickly getting an overview about a new field and summarizing a research paper to reviewing yourself and detailed citation analyses, scienceOS helps you save time and implement AI into your scientific workflows.

Start exploring scienceOS for free and see how it can support your research.