One context engine, several AI interfaces
How scoped retrieval, grants, and a shared MCP gateway let assistants use the same underlying evidence without inheriting the same permissions.
Engineering
Publishing is a system, not a button
Drafts, URLs, and the quiet work of keeping a public record coherent.
How we test an installable desktop release
Our evidence ladder for source tests, installable candidates, native platforms, and live connector checks.
Connectors are synchronization contracts, not import buttons
How Zortex reads API and local sources, records revisions, and distinguishes missing information from failed synchronization.
The cost of keeping personal context fresh
A small model of API calls, changed records and local storage shows where personal-context refresh gets expensive, and where it does not.
From source records to a world model
The mechanics behind event history, identity resolution, evidence-linked claims, and a graph that can be rebuilt.