Upload a document.
Your AI knows it in seconds.
No manual Q&A pairs. No training data. Upload your PDFs, docs, and manuals — Flux chunks, embeds, and indexes them automatically. Your AI answers from real business information.
HOW IT WORKS
From document to answer in four steps.
Upload
Drop a PDF, DOCX, Markdown, or plain text file. Or paste a URL.
Parse & Chunk
Text extracted, split into ~512-token chunks with overlap for context continuity.
Embed & Index
Each chunk embedded via bge-base-en-v1.5 and stored in Cloudflare Vectorize.
Retrieve & Answer
Visitor asks a question → vector similarity search → grounded answer with source.
KNOWLEDGE-GROUNDED
Answers from your data. Not the internet.
Flux retrieves the most relevant chunks from your knowledge base, injects them into the AI's context, and generates an answer grounded in your actual documentation. When confidence is low, Flux admits it doesn't know — no hallucinations.
Source citations
Answers reference the specific article they came from.
Confidence scoring
Low-confidence queries escalate to a human operator.
Fallback search
If vector search returns nothing, Supabase keyword search kicks in.
AUTO GAP DETECTION
Your AI tells you what it doesn't know.
Every time Flux can't answer confidently, it logs the gap. A daily AI job clusters these gaps into topics and generates draft articles for your team to review and publish.
Most knowledge bases only get smarter when someone manually adds content. Flux actively identifies what's missing and writes the first draft for you.
SELF-SERVICE
Published articles become a help center.
Mark any article as "Show in Help Center" and it appears on the widget's home screen as a searchable FAQ. Visitors find answers without starting a conversation — reducing conversation volume automatically.
SUPPORTED FORMATS
Bring your docs. All of them.
Each format is parsed natively — PDFs via unpdf, DOCX via mammoth, Markdown and text directly. URLs are crawled and extracted. All stored in R2 with per-tenant isolation.