# Enterprise RAG & Document Retrieval | iKB

> How iKB searches documents, embedded files, data records and approved connections, keeping calculated results apart from document evidence.

Source: https://kbmy.ai/architecture/data-retrieval

[iKB](https://kbmy.ai/) / [Architecture](https://kbmy.ai/architecture) / Data & retrieval

# Enterprise RAG & Document Retrieval How iKB finds supporting sources.

How iKB searches documents, embedded files, data records and approved connections, keeping calculated results apart from document evidence.

[Request a demo](https://kbmy.ai/contact?interest=Data%20%26%20retrieval)

[_Screenshot: iKB Topics Management showing sample corporate services and business operations knowledge bases._](https://kbmy.ai/assets/neo2/topics.webp) Connected knowledge iKB / Sample data

How it works

## Three steps to get started.

- 01 ### Process the sources Use uploads, supported public-link imports or approved connections. Check file and import status. Keep document versions, links to parent files and data types. The current version stays searchable while a replacement is processed.
- 02 ### Search permitted information Apply the topic's search method, source permissions and approved read actions. Flash mode and graph-based search have different capabilities.
- 03 ### Keep references Retain file references, transcript timestamps and calculation sources so a reviewer can check the answer.

Features & controls

## What data & retrieval includes.

### Import public shared files

Add a supported public HTTPS file link without first downloading and uploading the file yourself. Import methods include Google Docs, OneDrive, SharePoint, Dropbox and public repository files. Administrators must enable public-link imports and grant web-crawl permission. Public access, download permissions, file checks and processing limits apply. Provider methods have offline test coverage; confirm your links during evaluation.

### Knowledge inside embedded files

Search supported attachments inside Office documents, PDFs and email files when enabled. Keep each child file linked to its parent so users can trace a source. File types, nesting depth and processing limits apply. Access, retention and deletion controls cover the document family.

### Structured JSON and exact lookup

Look up exact fields in supported JSON data records. Keep data types, nested relationships and labelled paths, so a record can be checked alongside document sources.

### Controlled routing and refresh

Choose which documents a topic searches first. Set fallback and disclosure rules. Keep the current version searchable while a replacement is processed.

### Precise, source-backed retrieval

Combine keyword and meaning-based search, split multi-part questions and rank the results. Check that the answer is supported by the sources. Keep exact phrases, section names, definitions, table names and page references. Link individual claims to supporting evidence.

### Flash retrieval

Use Flash mode for straightforward document questions. It normally runs one embedding, one authorised vector search and one source-based answer call. With follow-up anchoring enabled, eligible short questions can add a second embedding and search using recent conversation context. The contextual result is used only when its search score improves enough. This adds no answer-model call. Flash topics skip dedicated decision tasks, analytics and tools.

Architecture and controls depend on your deployment. Test source quality, permissions, privacy, model failures, retention and exports against your requirements.

Continue exploring

## More from architecture.

[Decision Records](https://kbmy.ai/architecture/decision-records) [Deployment options](https://kbmy.ai/architecture/deployment)

[View the architecture overview](https://kbmy.ai/architecture)

## Test iKB on your team's questions.

Bring sample sources and clear requirements. Review the answers and controls before rollout.

[Book a demo](https://kbmy.ai/contact)