Agents overview

Bundata prepares your document base so agents can deliver accurate, grounded answers with source lineage and extraction confidence. Use the Vector Catalog for retrieval, structured extraction for entities and tables, and metadata for filtering and traceability.

What agents need from Bundata

• RAG enablement — Clean, contextually aware chunks and embeddings in the Vector Catalog so agents retrieve the right context.
• Structured data — Extracted entities, tables, and metadata in a consistent schema for reasoning and citations.
• File metadata enrichment — Source details, timestamps, and semantic tags so agentic systems can make confident, traceable decisions.

When to use Bundata for agents

• AI copilots — Customer support, legal review, R&D assistants that need to reason over your docs.
• Multi-document reasoning — Agents that answer questions across many documents.
• Internal assistants — Onboarding, HR, or ops agents that rely on internal knowledge bases.

How it fits with the rest of Bundata

• Extraction — Produces the smart bites and metadata that agents consume.
• Vector Catalog — Stores indexed content; agents query it via Vector Search or the search API.
• Workflows — Can trigger agent runs or post-process extraction output for agent use.

Grounding from catalog

Agents ground answers by retrieving relevant smart bites from the Vector Catalog, then passing those chunks to the LLM. Bundata ensures chunks have metadata and source lineage so responses can be cited and audited.

Common mistakes

• No metadata filtering — Use metadata to scope retrieval (e.g. by department, date) so agents don’t pull irrelevant context.
• Skipping source lineage in UI — Always show which document and run produced the grounding chunks.
• Treating extraction as one-off — Keep the catalog fresh with scheduled workflows so agent context is up to date.

Next steps

• Knowledge assistants — Patterns for doc-based assistants.
• Grounding from catalog — How to retrieve and pass context to the LLM.
• Vector Search overview — Semantic retrieval for RAG.