App Workflows

User journeys inside the application.

Chat with Documents

User navigates to /chat
Empty state shows "Ask your documents" prompt
User types a question and presses Enter (or clicks Send)
System runs agentic retrieval: intent classification → query planning → vector search → reranking → evidence validation
Answer streams in token-by-token via SSE
Inline [1], [2] citation links in the response are clickable
Citation panel slides open showing: doc title, section, page, excerpt, download link
Retrieval badges show: sources used, queries generated, latency
Multi-turn: follow-up questions use the same conversation context
Conversational messages (greetings, thanks) are answered directly without retrieval
See: Agentic Retrieval & Chat, Chat UI

User navigates to /upload
Drops or selects files in the dropzone
Client validates file size (max 100MB) and type
Progress bar shows per-file upload status
On success: toast notification, green checkmark
RAG processing status appears: "Processing — chunking, classifying, embedding..."
Text documents (PDF, TXT, CSV, Markdown, JSON) are automatically chunked, classified, summarized, and embedded into the vector store
Non-text files (images, audio, video) are uploaded to B2 but skip RAG processing
On failure: red status icon with error message
User can clear completed uploads
See: File Upload, Document Pipeline