Skip to main content

Knowledge Base

The Adeptia Knowledge Base lets you store, organize, and search documents so that AI-assisted integrations can draw on your own content.

What the knowledge base does​

At its core, the knowledge base is a managed document store with semantic search. You add documents to it, and the platform indexes them so that natural-language queries can retrieve the most relevant passages, ranked by how closely they match the meaning of the question rather than just the words used.

How documents are organized​

Documents in the knowledge base are grouped into collections. Each collection is a named, isolated workspace, and the platform scopes access so that a caller sees only the collections it is permitted to use. Within a collection, documents are further organized using a taxonomy path, a hierarchical label you assign when adding a document (for example, a category, subcategory, and topic). This structure lets you filter searches to a specific area of your content.

Documents can also be associated with a tenant category, which provides an additional dimension for organizing and filtering content across a shared collection.

Adding documents​

You can add documents to the knowledge base one at a time or in bulk. When adding a single document, you provide the file and a taxonomy path that places it in your content hierarchy. You may also supply a custom identifier for the document; if you do not, the platform assigns one automatically.

For large documents that would take too long to process in a single step, the platform supports an asynchronous ingestion path. You submit the document and receive a job identifier immediately; you can then check the status of that job at any time to see whether processing has completed, is still in progress, or encountered an error. Possible job states include queued, processing, completed, and failed.

For bulk uploads, you can submit a list of files in a single operation. The platform processes each file and reports back how many succeeded and how many failed, along with per-file details.

Supported document formats include PDF, DOCX, TXT, HTML, and Markdown, with a maximum file size of 25 MB per document.

Searching the knowledge base​

Queries use natural language: you describe what you are looking for, and the platform returns the most semantically relevant document passages, each with a relevance score. You can control how many results are returned and optionally narrow the search to a specific taxonomy path or tenant category.

When you have both a central shared collection and a customer-specific collection, you can search across both in a single query. When the merge path runs, each result indicates which collection it came from. You can also choose to search only the customer-specific collection when that is more appropriate.

Retrieving and managing documents​

Once a document is in the knowledge base, you can retrieve its full content, organized as the indexed chunks the platform created during ingestion. Each chunk carries metadata including its position in the document, the source filename, its category, and its taxonomy path.

You can list all documents in a collection, optionally filtered by tenant category, with pagination support for large collections. You can also see document counts broken down by category, which gives a quick picture of how content is distributed across your taxonomy.

To remove a document, you delete it by its identifier. Deletion removes the document and all of its indexed chunks permanently; the operation cannot be undone.

Preferences and rules​

Each collection can store customer-specific preferences and rules alongside its documents. Preferences are key-value settings that apply to the collection. Rules carry a type, a predicate, and an action, and can be enabled or disabled. You can read the preferences and rules for a collection, optionally filtering rules by type and paginating through large sets.

Providing feedback on search results​

After receiving search results, you can submit a signal indicating whether the response was helpful or not helpful. This feedback, tied to the specific query response by its identifier, feeds into the platform's content improvement process. You may also include a brief free-text explanation with your signal.

Checking service availability​

You can verify at any time that the knowledge base service is available and healthy. This is useful for diagnosing connectivity issues before running ingestion or search operations.