Architecture

Witzo uses a retrieval-based architecture so responses are grounded in your connected sources rather than relying only on a generic model response.

End-to-End Flow

1. A workspace connects website pages or uploads documents.
2. Witzo processes the source content into retrieval-ready chunks.
3. The indexed content is stored for future lookups.
4. A visitor sends a message from the widget.
5. The system verifies widget and usage state.
6. Relevant content is retrieved and passed into response generation.
7. The answer, conversation event, and any lead metadata are recorded.

Ingestion Pipeline

The ingestion layer is responsible for turning raw source material into structured knowledge:

• website crawling and readable-text extraction
• document parsing for supported file types
• content normalization
• chunking into smaller retrieval units
• indexing so relevant segments can be found later

Retrieval and Answer Generation

At runtime, Witzo:

• validates the widget and workspace context
• checks conversation availability
• retrieves relevant content from indexed sources
• generates a grounded answer from the retrieved context
• logs the conversation outcome for reporting and review

Workspace and Tenant Boundaries

Witzo is organized around workspaces so settings, content sources, and operational data can be managed at the workspace level.

This means teams can:

• keep one workspace separate from another
• control who can edit settings and data sources
• review leads and conversations within the intended operational boundary