AI chat & search
Chat is a retrieval-augmented (RAG) experience: your question is matched to relevant chunks from your indexed library, then an AI model composes an answer using that context.
Semantic search
Unlike keyword search alone, embeddings capture meaning—so questions like “How do we handle timeouts?” can match runbooks even if they never use the word “timeout” verbatim. Quality depends on chunking, model choice, and how cleanly text was extracted from your originals.
Citations
When possible, the UI surfaces which documents or sections informed the answer. Always verify critical facts against the source document—models can misread tables, code, or version-specific paragraphs.
Writing better questions
- Be specific: name a product area, API, or error string.
- Ask one thing at a time; follow up in a new message if needed.
- Mention environment (“production”, “staging”) if your docs distinguish them.
- If answers feel generic, upload more targeted docs or split huge PDFs into smaller files.
Limits & privacy (high level)
Inference runs in our Cloudflare Workers environment using your org-scoped index—your files are not used to train public foundation models through this product path. For legal or compliance wording, use your order form and DPA; this documentation is descriptive, not contractual.