Tiledesk offers a powerful semantic and hybrid RAG (Retrieval-Augmented Generation) feature – the Knowledge Base – purpose-built to deliver accurate, context-aware responses based on your organization’s knowledge.
RAG
The RAG approach combines generation with retrieval. This means that before answering, the system searches for relevant content – from internal documents, online articles, product manuals, or structured databases – and then injects that context directly into the prompt sent to the generative model (LLM).
Hybrid search: semantic & full-text, combined
Tiledesk supports both full-text and semantic search, and combines them in a hybrid strategy to maximize recall and precision:
- Full-text search ensures that keyword-based matches are quickly retrieved.
- Semantic search leverages vector embeddings to understand the meaning of the user’s request, even when the exact terms are not used.
This hybrid retrieval mechanism is particularly effective in scenarios where language is ambiguous or the information is highly specific and domain-dependent – such as technical support, compliance, healthcare, finance, or manufacturing.
Once the most relevant data is retrieved, Tiledesk augments the prompt with that content, enabling the LLM to:
- Provide fact-based answers grounded in your own data
- Reduce hallucinations and model drift
- Deliver domain-specific responses aligned with your business terminology and tone
- Answer in multiple languages, based on the underlying data source
The engine supports multiple contents formats and data sources allowing you to:
- Upload structured and unstructured content (PDFs, DOCs, HTML, Markdown, HTML etc.)
- Connect dynamic sources through APIs or webhooks
- Update content in real time or via scheduled syncs
This makes it ideal for organizations that want to maintain full control over the knowledge used by their AI Agents and ensure that responses are always aligned with company policies, updated documentation, or evolving product specifications.
The Ask to Knowledge Base action allows you to query a knowledge base in a flow, offering advanced control over every step of the process. You can select which Knowledge Base to query, tune the AI by configuring parameters such as the hybrid search percentage (to balance semantic and full-text search), the LLM model to use, the temperature to control response creativity, and the system context to set the assistant’s role.
