Supported files
The application accepts TXT, MD, CSV, JSON, PDF, DOCX, PPTX, XLSX and PNG/JPG/WEBP. Readable text is extracted from plain text, Word, PowerPoint and PDFs. Images and XLSX files are stored, but their content is not currently extracted for retrieval. Scanned PDFs require OCR that is not implemented here.
Limits and retrieval
Each file is limited to 8 MB; a company can have up to 100 files. Extracted text is capped at one million characters per file. Text is split into chunks; retrieval stays limited to selected company-owned evidence. Configured NVIDIA/Azure embeddings are stored beside chunks and used for hybrid semantic/keyword retrieval. Worker indexing continues in bounded batches; failures retain keyword retrieval.
Use in a Pod
Select only the files the Pod should consider. The worker receives excerpts from those files and the saved company blueprint. Retrieved content is evidence, not governing instructions. Review the resulting citations, assumptions and unresolved questions before approval.
Ownership
Files are company-scoped. The API checks that a selected knowledge ID belongs to the company. Removing a company deletes its associated documents, Pods and results.