Changelog
Only things a user can notice make this list. The pace is the point — this is what beta means here.
2026-08-04
When any agent searches a brain and gets nothing back, that query is now added to the brain's exam automatically — and the next re-read of the sources chases it. Hitting a gap is the bug report. The whole loop is written up on /collective.
2026-08-04
A family's parent brain used to generate its exam blind to what the children hold, so it failed questions its own family could answer. Generation now sees the family — the same scope the exam searches. Scores jumped accordingly.
2026-08-04
Top-ups and purchases now settle over our own checkout: USDT, USDC and BTC across four chains, a QR on the payment page, round amounts. No processor between you and the balance.
2026-08-04
hi@mozg.sh is gone. Every page that pointed at email now opens /chat — a direct thread to the person who builds this, answered from the same screen the beta reports arrive on.
2026-08-04
A new brain_feedback tool lets any agent flag a note that reality contradicted — the report lands on the owner's dashboard, the note keeps answering until a human decides. Plus: the plugin now sniffs your repo at session start and points the agent at matching brains.
2026-08-04
Five real queries into any paid brain, free, straight from your agent — then it asks to be bought. Storefronts show the questions a brain passed on its own exam, and /pricing finally says out loud what costs money and what never will.
2026-08-04
Svelte 5, Tailwind v4, Vercel AI SDK, OWASP cheat sheets and ASVS 5.0, Anthropic's prompt-engineering tutorial — free. Slot Studio, Slot Animation Craft and Slot Art Direction — a working studio's paid knowledge, $19–29 once.
2026-08-04
Extraction moved to a faster model after an A/B on the same exam showed no quality loss, big pages extract their segments in parallel, identical text is never paid for twice, and interactive uploads jump the queue ahead of background refreshes.
2026-08-04
Paste a documentation URL — the whole site is discovered (GitHub tree, llms.txt, sitemap, or a link walk), read into notes, a goal is drafted from the material, and the exam runs itself. JS-only sites fall back to their source repository.
2026-08-04
Every check is judged three times, majority wins — two re-sits on unchanged material now agree. The scorecard names what stands between a brain and 100%, tagged add-material or deepen-notes.
2026-08-03
Balance top-ups (crypto now, cards on the way), one-time brain purchases with 95% to the author, and a price field at creation for both the web form and agents over MCP.