Skip to main content
The knowledge base gives your agent access to information it can reference during calls — product details, FAQs, pricing, opening hours, and more.

​
How it works

  1. You upload documents (text or files)
  2. TalkPilot automatically splits them into chunks and creates vector embeddings
  3. During a call, when the agent needs information, it searches the knowledge base
  4. The most relevant chunks are included in the agent’s context
This is called RAG (Retrieval-Augmented Generation) — the agent retrieves relevant information and uses it to generate accurate answers.

​
Adding documents

​
Manual text

  1. Go to agent detail > Wissensdatenbank (Knowledge Base)
  2. Click Dokument hinzufuegen (Add Document)
  3. Enter a title and paste your text content
  4. Click Speichern (Save)
The document is immediately processed and available within seconds.

​
File upload

Upload files by dragging them into the upload area or clicking to browse. Files are automatically parsed, text-extracted, and indexed.

​
Document status

After uploading, documents go through a processing pipeline:

​
Knowledge Base tool

For the agent to actually search the knowledge base during calls, a Knowledge Base tool must be configured:
  • Auto-activation: The tool is automatically enabled when the first document becomes ready
  • Auto-deactivation: The tool is automatically disabled when the last document is deleted
You can configure the tool’s behavior:

​
Managing documents

  • Edit: Click on a manual text document to update its content. Editing triggers automatic re-processing.
  • Delete: Remove a document and all its chunks/embeddings.
Via API: You can create and manage text documents programmatically. File uploads are Dashboard-only. See Knowledge Base API.

​
External integration (Direct Chunk API)

If you manage your own chunking and embedding pipeline (e.g., via n8n, LangChain, or a custom ETL), you can write pre-processed chunks directly to an agent’s knowledge base — bypassing TalkPilot’s automatic processing. This is useful for:
  • CRM / ERP sync — keep product data, pricing, or customer info always up-to-date
  • Custom chunking — control chunk size and overlap for your specific use case
  • External embedding models — use your own embeddings (must be text-embedding-3-small compatible, 1536 dimensions)
Typical flow:
For full endpoint documentation, request/response examples, and the chunk data model, see the Direct Chunk API reference.

​
Tips

  • Keep documents focused on one topic each — this improves search accuracy
  • Use clear, factual language — avoid ambiguous phrasing
  • Update documents when information changes (pricing, hours, etc.)
  • Check document status after upload to confirm successful processing