Skip to main content
Apragya AI
Docs · AI & Automation

Feed your agent context with RAG and Knowledge Bases

Upload your team's docs to a Knowledge Base, attach it to an agent, and let the agent retrieve relevant context before generating.

Retrieval-Augmented Generation (RAG) lets an agent reach for your team's own knowledge - product docs, runbooks, policy PDFs, past tickets - instead of relying on what the base model happens to know. The platform's Knowledge Bases module handles the boring parts: ingestion, chunking, embedding, vector store, retrieval API.

Create a Knowledge Base

  • Sidebar > Suites > Solution > Knowledge Bases > New KB
  • Pick the scope: tenant-wide (any agent can attach), app-scoped (only agents inside the named app), or agent-scoped (the KB belongs to one specific agent)
  • Set a chunking strategy - the default works for most prose. Tighten for legal or structured data where paragraph boundaries matter; loosen for narrative content where broader context helps.

Ingest content

  • PDFs, DOCX, Markdown, HTML, plain text - drag and drop
  • Web URLs - paste a list; the platform fetches and parses
  • Sitemap URL - the platform crawls and ingests the entire site
  • Manual paste - text box for one-off content blocks
  • S3 bucket, Google Drive folder, Notion workspace - via connectors

Attach to an agent

Agent Builder > Knowledge Bases tab > Add. Once attached, the agent gets an implicit kb.search tool granted for that KB. The system prompt should mention the KB and when to query it; the model handles the rest.

Retrieval modes

  • Auto (default) - the agent decides when to query the KB based on the prompt
  • Always-on - every model call fires a retrieval against the KB first; useful for tightly scoped helpdesk agents
  • Manual - the agent only retrieves when it explicitly emits a kb.search tool call; useful when you want fine-grained control
RAG queries consume AI Credits at a lower rate than a standard model call, because the heavy lifting - indexing your content - happens up-front at ingest. Query cost stays predictable no matter how large the knowledge base grows.

Updating a KB

Re-ingest is delta-based - only changed chunks re-embed. Add/remove/replace individual sources from the KB's Sources tab. Bulk re-ingest is available via the SDK for scheduled refreshes (e.g., nightly Notion sync).

Need help building this? Talk to a real engineer

← All docs
Ask Vippy anything