Bot Knowledge
Open Bot Knowledge
After purchasing Bot Knowledge, open Configurations → Bot Knowledge in the dashboard sidebar. This tab contains scanned-PDF enhancement, document upload, your bot's namespace, and implanted documents.
Supported Files and Limits
ProxyAI accepts DOCX, ODT, RTF, PDF, PPTX, XLSX, XLS, CSV, EPUB, ZIP, HTML, HTM, RSS, XML, JSON, TXT, Markdown, TeX, and LaTeX files.
Each file can be up to 10MB, and must come to no more than 4MB of text once we extract it. Most documents are far under that, a long PDF is mostly images and layout, and only the text is indexed. If a document is too large, split it into smaller files. There is no fixed total per bot, storage is billed monthly instead of capped, so the practical limit is your credit balance.
A separate ceiling applies to extracted text: roughly 10 million characters per document. Compressed formats can expand far beyond their file size, a modest spreadsheet or archive can produce tens of millions of characters, so a small file can still exceed it. Split oversized sources into several documents.
Upload and Review
- Drag files into Knowledge Document Upload or choose files from your computer.
- Wait while ProxyAI extracts content.
- Review each document before implantation.
- Confirm only current, relevant content is included.
- Implant the reviewed document so it becomes searchable by the bot.
Prefer concise source files with clear headings. Remove duplicate or obsolete versions to reduce conflicting retrieval results.
Enhance Scanned PDFs
Enable Enhance scanned PDF content when a PDF contains pages without a text layer. Your browser downloads an OCR model and processes those pages on-device; documents are not sent to an OCR service.
Leave this option off for PDFs that already contain selectable text. OCR adds processing work and cannot guarantee perfect extraction, so review output before implantation.
Manage Documents
Implanted documents lists files embedded and searchable by the bot. Each entry shows its name and size. Choose Remove to stop using a document.
The displayed Namespace belongs to that chatbot. Upload documents through the intended bot's Bot Knowledge tab rather than reusing another bot's ID or namespace.
Documents with the same filename are setup as a replacement and update of the knowledge, use an unique naming for your documents for any other intentions.
Test Grounded Answers
Ask questions directly covered by implanted content, then ask related questions not covered by it. Verify answers follow source material and do not rely on obsolete documents. Use Usage Log to review Knowledge implant, retrieval, and storage activity.
Where Your Documents Are Stored
Implanted documents are held in a managed vector database. Each chatbot gets its own namespace in that index, and a search is scoped into an isolated space, so one bot cannot retrieve another bot's content.
One thing that stand out is our searches are hybrid. Every piece of data is indexed twice, once semantically and once as keyword terms. A question is matched both ways and the two result sets are merged, so query handles paraphrased questions and exact product codes or part numbers equally well. Accuracy of this arrangement is higher than any indexer product out in the market.
Removing a document deletes its data from the namespace. Deleting your account, deletes the whole namespace.
Storage Billing and Retention Policy
Bot Knowledge bills two things against your credit balance, both shown in Usage Log:
- Requests, charged when documents are indexed and each time the bot retrieves. Because the index is hybrid, one chunk costs two requests (semantic and keyword), and one retrieval also costs two.
- Storage, charged monthly for what your namespaces hold.
Storage is settled on the first of each month. If your balance does not cover the fee, the charge is still applied and your balance goes negative.
A negative or zero balance stops your bots from replying. Chatbots require a positive credit balance to answer, so once the balance goes under, visitors receive an "assistant is currently unavailable" message until you top up. This is the same behaviour as running out of credits for any other reason.
Separately, your knowledge bases enter a 30-day notice period and we email you. Nothing is deleted during that period, your documents stay indexed and are used again as soon as your balance is positive.
To end the notice period, top up enough to bring your balance back to zero and cover the outstanding amount that built up while you were behind. A partial top-up that leaves a balance below what is owed does not lift the notice.
If the balance is still outstanding after 30 days, the stored documents are deleted and we email you to inform. Your bots, settings, and conversation history are unaffected: once your balance is positive they answer normally again, just without your uploaded documents. Deleted documents cannot be recovered; restoring a knowledge base means uploading the files again.
To avoid reaching that point, enable low-balance alerts in your account settings.