ScienceOL Docs
  • English

      • Knowledge Bases

      Knowledge bases allow agents to access your documents and data during conversations. Upload files, and agents will retrieve relevant content to provide informed, grounded responses.

      Overview

      A knowledge base is a collection of documents that an agent can search and reference. When a user asks a question, the agent uses retrieval-augmented generation (RAG) to find relevant passages from the knowledge base and incorporate them into its response.

      This is useful for:

      • Company documentation and internal wikis
      • Research papers and literature
      • Product manuals and FAQs
      • Any domain-specific content the agent should know about

      Creating a Knowledge Base

      1. Navigate to the Knowledge section in the sidebar
      2. Click Create Knowledge Base
      3. Give it a name and description
      4. Start uploading files

      You can also create knowledge bases via the API:

      curl -X POST /scilaxy/api/v1/knowledge-sets \
  -H "Authorization: Bearer {token}" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "Product Docs",
    "description": "Product documentation and guides"
  }'

      Uploading Files

      Supported file formats:

      • Documents — PDF, DOCX, TXT, Markdown
      • Data — CSV, JSON
      • Code — Source code files

      Files are processed upon upload: text is extracted, chunked, and indexed for retrieval.

      Upload via the web interface by dragging files into the knowledge base, or via API:

      curl -X POST /scilaxy/api/v1/knowledge-sets/{id}/files \
  -H "Authorization: Bearer {token}" \
  -F "file=@document.pdf"

      Linking to Agents

      To give an agent access to a knowledge base:

      1. Open the agent configuration
      2. In the Knowledge Base section, select a knowledge base
      3. The agent now has access to the knowledge retrieval tools

      Each agent can be linked to one knowledge base. The agent automatically gets access to the knowledge_read and knowledge_write tools.

      How Retrieval Works

      When an agent needs information from the knowledge base:

      1. The agent calls the knowledge read tool with a search query
      2. The system performs similarity search across the indexed documents
      3. The most relevant document chunks are returned to the agent
      4. The agent incorporates the retrieved content into its response

      The agent decides when to search the knowledge base based on the user's question and the conversation context. It can also write new entries to the knowledge base to store information for future retrieval.