How Unstructured search works

Unstructured search creates a search index over documents from a configured source. Supported sources include Amazon S3, Google Cloud Storage, and Google Drive.

When you ask a question, Unstructured search retrieves relevant document sections, called chunks. It passes these chunks to Reltio-provided AI models to generate an answer grounded in your organizational knowledge.

Before indexing the full document source, you can evaluate the retrieval configuration with a sample document. In the test playground, you can ask questions, review answers and retrieved chunks, and confirm that the configuration returns relevant content. You can then apply the configuration to the complete source and create the knowledge base index.

Key Unstructured search concepts

Source

The configured Amazon S3, Google Cloud Storage, or Google Drive location that contains the documents.

Retrieval configuration

The search method and number of results returned for a query. Unstructured search supports semantic search and hybrid search, which combines semantic and keyword retrieval.

Search index

The stored, searchable representation of the processed document corpus. It contains the information Unstructured search uses to locate relevant chunks efficiently.

Retrieved chunks

Sections of document content that Unstructured search identifies as relevant to a question. Unstructured search retrieves chunks instead of sending every complete document to an agent for each question.

Test playground

A pane on the Create knowledge base page for evaluating retrieval quality against a sample document before creating the complete knowledge base. It displays generated answers and the chunks retrieved for each test question.

Indexing schedule

The schedule for processing documents from the configured source: Manual, Daily, or Weekly.

When to use Unstructured search

Use Unstructured search when users or agents need to answer questions from approved enterprise documents or authoritative knowledge sources of an organization. Common uses include:

• Searching policies, contracts, reports, and technical documentation.

• Creating an internal knowledge assistant for organizational processes and systems.

• Helping support teams find information from product documentation and troubleshooting material.

• Answering questions from product manuals and other operational documents.

When used with custom agents, Unstructured search also supports data unification scenarios such as:

• Combining customer profile context with product documentation and support runbooks.

• Combining governed entity data with policies, contracts, and regulatory documents for compliance reviews.

• Using customer or vendor context together with onboarding documents to answer operational questions.

Unstructured search keeps answers grounded in the selected document corpus instead of relying only on general model knowledge.

Knowledge bases in Custom agents

Custom agentscan use one or more Unstructured search knowledge bases. The agent author selects an accessible Unstructured augmentation tenant and chooses the knowledge bases available in that tenant.

During testing and conversations, the agent retrieves relevant document content and uses it to answer questions according to its system prompt and other enabled capabilities.

To get started, see Add Unstructured search knowledge bases to a custom agent and Create, test, and submit an agent for review.