Knowledge Source

Sourcesible Product Documentation · Insights

Overview

Knowledge Source lets you upload reference documents or any unstructured content that AI Agents can retrieve from using semantic search. Unlike Datasets which are queried row-by-row, Knowledge Sources provide narrative context that shapes the quality and relevance of agent responses. Once indexed, a Knowledge Source can be connected to one or more AI Agents.

Creating a Knowledge Source

Set Up the Knowledge Source

In the left sidebar, under Insights, click Knowledge Source. Click + Create Knowledge Source in the top-right corner. On the Set up Knowledge Source page, enter a Knowledge Source Name (maximum 50 characters). Enter a Description to describe what this knowledge source contains and how the agent should use it this description is shown to the model during inference to help it choose the right source. Click Create Knowledge Source → to proceed.

Upload Files

After the knowledge source is created, the file upload area appears. In the Drop Files section, click Choose file to Upload or drag and drop a file directly into the upload area. Select your file from the file picker.

Supported file formats: pdf, pptx, docx, html, xlsx, csv, txt.

The file appears in the Files list with its name and file size. Click Save.

Monitor Indexing Status

Once saved, the file's Embed Status will show Processing while Sourcesible indexes the content. When indexing is complete, the status updates and the knowledge source becomes available to connect to AI Agents. The file list shows File Name, File Size, Embed Status, Created by, and Create Date for each uploaded file.

Connecting a Knowledge Source to an AI Agent

After creating a Knowledge Source, connect it to an agent from the AI Agents configuration page. Under Knowledge and Data Access → Knowledge Sources, click + Add Knowledge, select the knowledge source from the picker, and review the description. Click Save Changes to apply.

For guidance on writing effective Knowledge Source descriptions and priority rules, see the AI Agents documentation.

Next Steps

Connect to an AI Agent Open AI Agents, navigate to Knowledge and Data Access → Knowledge Sources, and add the knowledge source.

Test retrieval quality Use AI Assistant with the connected agent to verify that the agent retrieves and applies the knowledge source content correctly.

Add more files Return to the Knowledge Source at any time to upload additional documents and click Save Changes.

Tips and Troubleshooting

Embed Status Stays on Processing for a Long Time

Symptom: After uploading a file and saving, the Embed Status column shows Processing for more than a few minutes.

Cause: Large files or files with complex formatting (e.g., multi-sheet xlsx, image-heavy PDFs) take longer to index. Transient processing delays can also occur under high system load.

Fix: Wait up to 10 minutes before taking action. If the status does not update, try deleting the file and re-uploading it. If the issue persists with a specific file, verify the file is not corrupted and is within supported size limits.

Agent Does Not Retrieve Knowledge Source Content

Symptom: After connecting a Knowledge Source to an agent, the agent still gives generic responses and does not appear to use the uploaded documents.

Cause: The Knowledge Source may still be indexing, or the agent's Instructions do not include a directive to use connected knowledge sources, or the Description is too vague for the model to know when to retrieve from this source.

Fix: Confirm the Embed Status shows a completed state (not Processing). Then update the agent's Instructions to explicitly instruct the model to use connected knowledge sources. Also update the Knowledge Source Description with a clear priority rule for example, "Prioritize this knowledge when responding to marketing strategy or campaign planning queries."

Unsupported File Format

Symptom: An uploaded file is rejected or does not appear in the Files list after uploading.

Cause: The file is not in one of the supported formats: pdf, pptx, docx, html, xlsx, csv, txt.

Fix: Convert the file to a supported format before uploading. For image-only files (e.g., scanned PDFs with no text layer), run OCR to extract the text before converting to a supported format.