Knowledge Base
The knowledge base gives your agent access to your content through Retrieval-Augmented Generation (RAG). Upload documents, and your agent can search and reference them during conversations to provide accurate, grounded responses.
How It Works
Upload Content
Add documents to your project's knowledge base. Supported formats include plain text, PDF, DOCX, CSV, and web URLs. You can upload files through the dashboard or via the API.
Automatic Processing
Conduet automatically breaks each document down into smaller passages and indexes them for semantic retrieval. You don't need to configure chunking, embeddings, or a vector store — it's all handled for you.
Semantic Search
When a user asks a question, Conduet finds the passages that most closely match the meaning of their question — not just keyword matches — and surfaces the best ones, ranked by relevance.
Context Injection
Retrieved chunks are injected into the agent's context alongside the user's message. The LLM uses this context to generate accurate, source-grounded responses.
Supported Content Types
| Type | Format | How to Add |
|---|---|---|
| Plain Text | Direct text input | Paste in dashboard or POST via API |
| .pdf files | Upload via dashboard or API multipart form | |
| Word Document | .docx files | Upload via dashboard or API multipart form |
| CSV | .csv files | Upload via dashboard or API multipart form |
| Web URL | Any public URL | Enter URL in dashboard or POST via API |
Managing Documents
The Knowledge Base page in your project dashboard shows all documents with their status, chunk count, and creation date. From here you can:
Add documents — Upload files, paste text, or enter a URL to crawl. Documents are processed asynchronously; the status changes from “processing” to “ready” when chunking and embedding are complete.
Test search — Use the search input on the Knowledge Base page to test semantic queries against your documents. This helps you verify that the right content is being retrieved for different types of questions.
Delete documents — Remove a document and all its embedded chunks. This immediately removes the content from search results.
Using Knowledge Base in Agents
There are two ways to connect your knowledge base to your agent:
Agent Step (Automatic)
Turn on the knowledge base toggle on an agent step and the agent will decide for itself when to look something up — perfect for open-ended questions. This is the recommended approach for most agents.
KB Search Step (Explicit)
Drop a Knowledge Base Search step into your workflow when you want full control over the query, how many results to pull, and how confident the match should be before using it.
Best Practices
Keep documents focused. Smaller, topic-specific documents produce better search results than large, multi-topic ones. Split broad documents into sections.
Use descriptive document names. Names help you organize and identify content in the dashboard. They also appear in search result metadata.
Test with real questions. After uploading content, use the search test feature to verify your agent can find the right answers. Adjust content if results are not relevant.
Keep content up to date. Delete and re-upload documents when your source content changes. Stale information leads to incorrect agent responses.