AI Knowledge

AI Knowledge lets AI agents retrieve information from unstructured documents and use it in downstream tasks. By ingesting internal company documents and product resources, the system transforms them into searchable knowledge that is accessible through natural language interactions, such as the AI Assistant. This supports day-to-day questions and more complex workflows where fast access to organizational knowledge matters.

What AI Knowledge Does

AI Knowledge helps you ingest documents into a knowledge store, run semantic search over the ingested content, and use the retrieved snippets as input for the AI Assistant.

It currently works with plain-text content provided as:

• Markdown files
• Text files

If your source content is stored in another format, extract the text first and convert it to one of the supported formats before ingestion.

How It Works

When users ask questions that relate to organizational knowledge, the AI Assistant searches ingested documents and uses matching content to ground its answers.

What users can expect:

• Workspace-level documents are prioritized when searching from a workspace.
• Answers include citations such as the document title or filename, and page numbers when available.
• The assistant only cites documents it actually used.

Permissions

Required permissions:

• Create, upsert, patch, delete: WRITE_KNOWLEDGE_DOCUMENTS
• Search, list, get: READ_KNOWLEDGE_DOCUMENTS

These permissions follow the standard workspace permission hierarchy. For example, MANAGE implies write, which implies read.

Ingestion

When you ingest a document, GoodData splits the content into chunks, creates embeddings for those chunks, and stores them for retrieval.

Default settings:

• Chunk size: 1000 characters
• Chunk overlap: 200 characters
• Split priority: ## headers, ### headers, #### headers, paragraphs, lines, sentences, words, characters

If you provide pageBoundaries in the request, chunks track which page numbers they span.

Document Scope and Visibility

Documents can be scoped to a workspace. When searching from a workspace, workspace-level documents are prioritized.

Knowledge REST API

Base path: /api/v1/actions/workspaces/{workspaceId}/ai/knowledge

Create a Document

POST /documents creates a new document.

Request fields include:

• filename (required)
• content (required): full document text
• pageBoundaries (optional): character offsets for page breaks
• title (optional)

The response includes numChunks.

Upsert a Document

PUT /documents creates the document if it does not exist, or updates it if it does. On update, mutable fields such as title, scopes, and content are updated and old chunks are deleted.

List Documents

GET /documents lists documents visible in the workspace. It supports cursor-based pagination.

Get, Patch, Delete

• GET /documents/{filename} returns document metadata.
• PATCH /documents/{filename} updates metadata without re-ingesting content.
• DELETE /documents/{filename} deletes the document and its chunks.

Search

GET /search runs semantic similarity search over document chunks. It returns results from the current workspace and excludes disabled documents.

The API supports:

• query (required)
• limit (default 10)
• minScore (default 0.0)
• scopes (optional OR-based filtering)

Topic tag filtering using scopes is available via the REST API, but it is not currently used by the assistant tools.

Common Use Cases

• Search internal documentation and policies from an AI assistant.
• Help users find the right metric, dashboard, or workflow by searching product documentation.
• Ground answers in organization processes, playbooks, and other internal knowledge.

Current Limitations

The following limitations apply in the current version:

• Content size limit is about 256 KB per ingestion request.
• Scope filtering is OR-only when multiple scopes are provided.
• Search is not paginated. It returns the top results up to the requested limit. Document listing is paginated.
• Only Markdown and text files are currently supported for ingestion. Support for additional file types may be added in a future release.
• Topic tag filtering is available through the REST API, but it is not currently used by the assistant tools.
• Organizations that opt out of data sharing cannot use AI Knowledge (data_share_opt_out=false).