Changelog
Every entry below is a real ship date from the build record. This is what the context layer for humans and their agents looks like being built: state, receipts, and no silent rewrites, in the product and in this page.
Provenance stamping shipped: each atom records whether it came from an email, a document, an AI session, or MrAgentˣ's own inference, plus which model wrote it.
- What you received and what you worked out with your AI never mix; the stamp is written by the database at creation and survives every later update.
- Where the source cannot be detected, the stamp stays honestly empty rather than guessed, and older atoms say unstamped instead of pretending.
The dashboard's heaviest pages were rebuilt to load only what the view needs.
- Long lists paginate at the database, so a year of history opens as fast as a week.
Full workspace export is live: every atom, every connection, every month of accumulated context, delivered as a signed download.
- Request an export from Workspace Settings and the queue handles it in the background; you get a private download link that expires after 48 hours, then the artifact is purged.
- Exports are metered by tier and one runs at a time per workspace, enforced by the database, not by politeness.
- Every new route shipped fail-closed: session required, admin gates on, access scoped to workspace members.
- The exit test is the point. Cancel everything tomorrow and the answer to "what do you keep" is: all of it.
All background intelligence now runs through one harness: the model is a bounded callee, handed one task and one prompt, identical across every client.
- No agent decides what happens next; the queue does, deterministically.
- This is the architecture that keeps loads flat and behaviour identical no matter which AI you connect.
Context Windows Close. AI Forgets Everything. Your Work Should Never Start From Zero.
MrAgentˣ is in private beta. Limited to the first 1,000 until launch.
Join the waitlist