Knowledge Base
A Knowledge Base is a collection of documents your agent can reference during conversations. Instead of relying solely on the LLM’s training data, agents with a knowledge base provide answers grounded in your information.
Why Knowledge Bases Matter
LLMs are powerful but imperfect. They hallucinate facts, go out of date, and know nothing about your specific business. Knowledge bases solve this.
How It Works
When a user asks a question:
- The agent searches the knowledge base for relevant content
- Matching passages are injected into the LLM context
- The agent responds using this grounded information
This is called Retrieval-Augmented Generation (RAG)—your agent retrieves before it generates.
Use Cases
- Product Support — Upload manuals, troubleshooting guides, warranty policies. Agents answer product questions accurately.
- Sales Enablement — Add pricing sheets, feature comparisons, objection handlers. Agents sell with confidence.
- HR & Onboarding — Store employee handbooks, benefits info, org charts. New hire questions handled instantly.
- Legal & Compliance — Include policies, regulations, approved language. Agents stay within bounds.
Key Concepts
One Agent, One KB
Each agent can link to one knowledge base via the globalKnowledgeBaseId field. This becomes the default source for all conversations with that agent.
KB Independence
Knowledge bases exist independently of agents. You can:
- Create a KB first, then link agents later
- Reuse one KB across multiple agents
- Update KB content without touching agent configuration
Document Types
Currently supported:
- PDF files (preferred) via the upload endpoint
- URL scraping (fetch and ingest page content)
- Text content (when the text upload endpoint is deployed, will be available soon)
