Document search: find company documents by meaning, with sources

Document search looks through the content of a workspace semantically, in the AI chat and in a dedicated search window, and cites the source for every result. It is the enterprise search of the Nexoria platform; what the term means in general is explained in the glossary: What is enterprise search?

This makes search the layer between storage and answer: the content comes from Nexoria Drive and from configured integrations, and the same index is also used by Nex, when it answers a question. Anyone who is only looking for a document uses the search window. Anyone who needs a fully formulated answer asks in the chat. The basis is the same in both cases.

Nexoria Drive, document search and the AI assistant Nex form the core of the Nexoria platform. Which components build on this is shown in the component overview.

When searching within a company, different wording can become an obstacle. The contract says notice period, but people search for exit clause. The minutes say delivery delay, but people search for postponement. A pure keyword search without suitable language processing or synonyms can miss such passages. In addition, the document must actually be part of the indexed and shared search content.

Source citations and document preview link the search result to the relevant passage. This lets your team verify results directly in the original and keep working with the knowledge found. It also makes it easier for new employees to get to grips with document collections that have grown over time.

Both searches have their place, but they start from different points.

Windows file search provides access to files via the taskbar and File Explorer, according to Microsoft. In File Explorer, for example, you select a drive or a folder; depending on the account and settings, cloud files also appear. In classic mode, the local search index covers Documents, Pictures, Music and Desktop by default. Which file contents are indexed depends on the file type and settings. On Copilot+ PCs, semantic indexing supplements conventional search.

Nexoria's document search searches the shared, indexed content of a workspace, i.e. the knowledge your team uses together. It finds content by its meaning, names the source document for every result with a preview of where the match was found, and searches only the workspace it is assigned to. On the same basis, Nex answers questions with source citations.

You keep the familiar route via File Explorer: on Windows, Nexoria Drive integrates into File Explorer as a drive. Shared content is indexed and is then available in document search.

• You know the file name or storage location and are searching on your own device: Windows file search takes you there.
• You know the content but not the file name, or you are searching your team's shared content: document search finds the right passage and cites the source.

A full-text search matches search terms against the indexed words of a document; depending on the system, word forms are normalised or stored synonyms are taken into account. A semantic document search converts text passages and the query into embeddings, i.e. numerical representations in which proximity stands for similarity of content, and in this way compares their meaning. A question about the exit clause can therefore make a paragraph on the notice period findable. How the method works in detail and where its limits lie is explained in the glossary.

For Nexoria document search, what counts is the indexed, shared content in Nexoria Drive and in configured integrations. You transfer files from SharePoint, an email inbox or a network drive, for example on a file server, into this content so that the search can find them. SharePoint, OneDrive and Google Drive can also be connected as knowledge sources on request. We set up the integration depending on the environment and permissions. On request, we also check whether further storage, CRM, ERP and in-house systems can be connected. Scope, updates and permissions are defined for each integration. How an integration works is described on the page Integrations with SharePoint, OneDrive and Google Drive. For a file server that has grown over time, the file server question in the FAQ explains the steps in detail.

Scanned documents and recordings are made searchable by the Speech and text recognition component; the recognised content goes into the same index. Who sees which results is determined by Workspaces: document search only searches the content of the workspace it is assigned to.

Start with a clearly scoped case, such as a project or a set of contracts, and clarify three points before transferring the content:

• Which shared documents are included? Record the origin, the authoritative document version and any missing documents. Check the indexing with your actual files.
• Who is allowed to see this content? Assign the test collection to suitable workspaces and check results and sources with different test accounts.
• Where will the originals continue to be maintained? Decide who transfers changes into the search content, so that answers are not based on outdated copies.

How to handle libraries, metadata and permissions before a switch is described in the SharePoint comparison. The following test tasks assume a test collection that has already been selected and shared.

Check content, document version and access with a small shared test collection. The following table contains four test tasks for your own trial.

For each task, record the question, expected source, actual result and document version used. A matching result is helpful; for important decisions, the source must also support the specific statement. AI answers can contain errors.

The test log with 20 test cases contains templates for source fidelity (case 10), permissions (case 16) and revocation of permissions (case 17). For case 16, compare the same question with two test accounts: one with access to the workspace and one without. Record the search results, preview and chat answer separately. For case 17, also note the time of observation after permissions are revoked.

The guide to introducing AI knowledge search brings these checks together into a pilot. What finding by content looks like in everyday filing is shown in the Example: finding the notice period without knowing the file name. For a test with your own documents, book an appointment.

Document search searches a workspace's content semantically and names the source for each result. This describes how the search works, not the quality of its results.

• Search works on the indexed content in Nexoria Drive and in configured integrations.
• How the workspaces are structured determines which content is searchable for whom.
• Scanned documents and recordings are made searchable by the Speech and text recognition component.
• The effort for the initial indexing of existing content depends on its size and is calculated individually during setup.
• The offer is aimed at business customers within the meaning of Section 14 of the German Civil Code (BGB); all prices are net prices.

You can find the current plans under Pricing.

The table shows, step by step, where document search processes which data.

Technical error monitoring and enabled add-on services such as Nexoria Mail, Nexoria Meet or the VoiceBot also have their own data flows; the service providers involved are listed in the processing register. Processing locations and safeguards for the entire platform are described on the page Security and privacy.

• Semantic search across the workspace's shared content, in natural language.
• Source citation for every result, with the document it comes from.
• Document preview in the search result, with tables, images and formatting, without downloading.
• Two access points: a dedicated search window outside the chat and search within Nex.
• Searchable content from Nexoria Drive and from configured integrations.
• Scanned documents and recordings searchable via text recognition and transcription.
• Separate knowledge base for each workspace, at both database and vector store level.
• No AI tokens consumed, so it remains available even when the allowance is used up.

Where this component fits into the overall picture is shown in the component overview.

What is compared is the method, not a specific product: a full-text search over a file repository versus Nexoria document search.

• Choosing enterprise search solutions
• Implementing knowledge management in a company
• Comparisons: SharePoint comparison · Confluence alternative · Nextcloud alternative