Sources

Sources are specialized knowledge bases that allow you to ground the AI Assistant in specific technical or proprietary data. By uploading documents to a source, the platform generates embeddings that enable the Assistant to retrieve and reference that information during research tasks.

How Sources Work

When you query the Assistant, it can pull context from these ingested materials via Retrieval- Augmented Generation (RAG). This ensures that findings are based on the latest research papers, internal reports, or specific technical documentation you provide.

Supported File Types

The platform currently supports creating knowledge bases from the following file formats:

• .pdf
• .md
• .txt

Data Processing and Embeddings

Once a file is uploaded, it must be processed to create the vector embeddings required for AI retrieval. This operation occurs in the background and may take some time depending on the file size and complexity.

• Processing Status: In the Sources list view, a "Processing..." tag indicates that the embeddings are still being generated.
• Assistant Availability: Information within sources in the processing state may not be available for the Assistant to reference until the status is cleared.

Managing Source Collections

The Sources page displays both system-provided knowledge bases and user-created collections.

System Sources

Some sources (such as "MRSA" or "Alzheimers") are provided by the system. These are read-only libraries maintained by administrators and cannot be modified or deleted by users.

User-Created Sources

You can build your own research libraries by clicking the New source button.

• Uploads: Add files by dragging and dropping them into the upload zone or by using the Upload files button.
• Metadata: Each source can be given a custom name, a detailed description, and tags to help categorize the information (e.g., Clinical, Proteomics, Literature).
• Modifying: Click Edit on a source page to update its name or description. You can continuously add new files to a source as your research evolves.

Privacy and Collaboration

Control over who can access your research materials is managed via the visibility toggle:

• Private: By default, new sources are private and only accessible to you.
• Published: Switching a source to "Published" makes it available to all other collaborators within the current project.

IMPORTANT

Remember that published sources are shared within the project context. Every collaborator in the project will be able to use the information in that source for their own Assistant conversations.

---

Ready to handle arbitrary research data? Let's move on to Datasets! 📊