20+ AI models. Expert agents that learn. Connect your tools. Private knowledge bases for teams. Not just another chat window.
Companies are right to be cautious about AI and sensitive information. Here's exactly what happens to your data — and why it's safer than you think.
When you upload a policy document or legislation extract, it's chunked into 1,200-character segments and stored in your Cloudflare R2 bucket — encrypted at rest, in a region you control. The original file never touches an AI model.
When someone asks a question, we retrieve the 3 most relevant chunks using vector similarity — roughly 3,600 characters of context. Only those chunks go to the AI model. The model never sees your full document. It sees fragments, stripped of structure, no more sensitive than what you'd paste into a search box.
Workspaces have owners, admins, and members. Members can query. Only admins can upload documents or invite others.
Every document upload, every query, every member change is logged with handle and timestamp. Exportable for compliance.
Workspace A's documents are invisible to Workspace B. Expert agents are scoped to their workspace. Queries cannot traverse boundaries.
Raw documents live in Cloudflare R2 object storage — not in any AI model's training data, not in any external vector database.
The AI model receives at most 4 chunks per query — roughly 4,800 characters of your content. Never a complete document.
All processing happens on Cloudflare's infrastructure — the same infrastructure trusted by millions of businesses worldwide.
Daily message caps exist to prevent exploitation and keep costs predictable — not to frustrate real users. Normal usage never hits the ceiling.
5 free turns to try it. No credit card required. Enter your handle at Stripe checkout — account unlocks automatically.
5 free turns. No card. Works in your browser, installs as a PWA on mobile.