Add files as knowledge
Upload supported documents and images, monitor processing, and verify that extracted content helps the agent.
Open Sources → Content. Select Files to upload material that is not available from a public URL. An agent can have up to 50 files. Each file must be 50 MB or smaller. Supported formats are PDF, DOC, DOCX, TXT, PPTX, XLSX, XLS, PNG, JPG, and JPEG.
Upload carefully
Choose files that are authoritative and readable on their own. A slide deck with sparse bullets or a scanned image can contain less useful context than a well-structured document. When you upload a batch, the system processes valid files in order. It skips invalid files. It stops the batch at the per-agent file limit.
The file list shows the status of each file: Pending, Processing, Extracting text, Creating embeddings, Completed, or Failed. Before you test a file, wait until it reaches Completed. Do not test a file while it still shows Extracting text or Creating embeddings.
Review and maintain
You can preview, download, or delete completed files. When possible, preview the extracted material. Then ask a known-answer question in Playground. If a new file supersedes an older policy, remove the old file. Otherwise the agent must reconcile contradictory facts.
The system indexes files when you upload them. The retrain worker refreshes website sources and enabled Notion sources. It does not refresh uploaded files. To update a file, upload the revised version. Verify the new result. Then remove the obsolete version. See Manage knowledge sources for the full source-selection workflow.