Project privacy

Updated October 24, 2026. This page summarizes what each project reads, sends, and stores.

• MCP tools: local tools return selected evidence to your AI client without calling a model or saving request/result content. The public MCP examples expose synthetic data. Your AI client's retention policy still applies.
• CodeArchitect web: bundled public source only; personal uploads and server-path selection are blocked. Questions/context go to OpenAI. This differs from the local MCP.
• OrcView web: questions and bundled source go to Anthropic. Configuration/history is private to a temporary browser session. SDK disk recordings require explicit opt-in.
• DemandMap: fictional records and offline vectors; uploads disabled. News research is available through the local workflow.
• Learning Accelerator (hosted backend retired; local software): per-browser temporary databases; notes/questions/PDFs go to OpenAI when AI features are used. Shared search is disabled; browser copies use sessionStorage.
• Markets (hosted backend retired; local software): selected indicators are available through the local dashboard; saved-data changes require operator authorization.
• AgenticETL and VariantGrid: trusted local operator tools that intentionally persist events. The examples use synthetic events.
• Brazil analysis and SQLHarness evidence: aggregate/research artifacts; source data and third-party publications retain their own terms.

Hosting providers may retain operational metadata. Some pages load external fonts/assets. AI providers apply their own data-handling policies.

Questions or deletion requests: andrewjorge.jobs@gmail.com.