Knowledge Base Q&A
Knowledge Base Q&A lets you upload your own documents into Adeptia and then ask natural-language questions against their content, receiving answers generated by a large language model grounded in your specific data.
What you can do
With Knowledge Base Q&A, you can build a private, searchable knowledge base from your own files and query it conversationally. This is useful when you need to extract insights, look up information, or automate question-answering workflows based on content that is specific to your organization, rather than relying on a general-purpose AI model alone.
You can query across everything in a knowledge base, or narrow a question to a single document within it. Both modes support conversational context, so you can pass in a previous question and answer to continue a multi-turn dialogue.
Supported document types
You can upload documents in a range of formats. The platform handles each format appropriately during ingestion.
| Category | Supported formats |
|---|---|
| Spreadsheets | CSV, XLSX |
| Images | JPG, JPEG, PNG, BMP, TIFF, HEIF |
| Other | DOCX, PPTX, HTML |
How documents are processed
When you upload a document, the platform stores its content as vector embeddings in a dedicated knowledge base. Each knowledge base is identified by a tag you assign, and each document within it is tracked by its file name. The platform splits document content into chunks before embedding, so that relevant passages can be retrieved efficiently at query time.
For PDF files, you can choose whether the platform reads the file directly as a PDF or first converts each page to an image before processing. The image-conversion path is automatically skipped for PDFs that exceed a configured page-count threshold, keeping processing practical for large files.
For reading content out of files, the platform supports multiple reader options depending on the file type, including a general-purpose document reader and an AI-powered document intelligence service for more complex layouts.
What you configure when uploading
When adding a document to a knowledge base, you provide or select the following:
| What you configure | Purpose |
|---|---|
| Knowledge base tag | Groups documents together under a single queryable collection |
| Document type | Tells the platform how to parse and load the file |
| PDF reader | Chooses how PDF content is extracted (direct reading or image-based OCR) |
| Image reader | Chooses how image files and image-converted PDF pages are read |
| Reader for other file types | Chooses the extraction method for formats such as DOCX, XLSX, and PPTX |
| Chunk size | Controls how the document is split before embedding, affecting retrieval granularity |
Checking upload progress
After submitting a document, you can check whether it has finished processing. The platform confirms when a document is fully indexed and ready to query, or reports that the upload did not complete successfully.
Querying your knowledge base
Once documents are indexed, you can ask natural-language questions. You choose whether to query the entire knowledge base or a specific document within it.
When asking a question, you can provide:
| What you provide | Purpose |
|---|---|
| Question text | The natural-language question you want answered |
| Chat history | A previous question and answer pair, enabling follow-up questions in context |
| Response structure | A JSON structure the answer should conform to, for use in downstream automation |
| Continue generating | Instructs the model to continue a response that was cut off |
| Include details | Requests that the response include source passage information alongside the answer |
| Send all data | Retrieves the full content of the knowledge base rather than only the most relevant passages |
Understanding the response
Every query response includes the model's answer along with a token count and a flag indicating whether the response was truncated. When you request additional details, the response also includes the source passage the answer was drawn from and the full conversation history.
If you request a structured response, the answer is returned as JSON conforming to the structure you specified, without additional explanation, making it straightforward to use the output in an integration flow.
Managing your knowledge base
You can remove content from a knowledge base at any time:
| Action | What it does |
|---|---|
| Delete a document | Removes a single document's embeddings from the knowledge base while leaving other documents intact |
| Delete a knowledge base | Removes the entire collection and all documents within it |