Streaming gets a refresh, Memory Vault becomes Your Memories
A week focused on what happens while the model is working — and on making memory feel more like a place than a vault. Plus a few new models and a much-needed sign-up cleanup.
Live thinking
When a model is reasoning through a hard problem, you can now see it happen. The activity panel streams the model's thinking text live as it works, so you're not staring at a spinner waiting for the final answer. It's a small change that makes longer responses feel substantially less opaque.
Streaming markdown, rebuilt
The streaming markdown renderer was overhauled. Code blocks now have richer chrome, tables render correctly under streaming pressure, and bold and italic spans no longer break across newlines. Output that used to occasionally glitch mid-stream now lands clean.
Memory Vault → Your Memories
Memory Vault is now called Your Memories. The change isn't just cosmetic — it reframes the surface from "the place we store stuff" to "the place that's yours." Same data, same encryption, friendlier home.
We also wired stored memories into Council Mode prompts. Unified answers now reflect what you've taught Anuma, not just what's in the immediate prompt.
Models
We added support for Claude Opus 4.7, made GPT 5.2 the standard-mode default, and migrated Qwen 3.6 Plus to a faster provider (Fireworks) for noticeably lower latency.
Audio Studio gets custom credits and downloads
Audio Studio now has custom credit costs that better reflect each model's actual usage, and you can download generated audio and video directly from the player toolbar.
Quality fixes
We chased down a multi-tab race condition that could lose data when chatting in multiple browser tabs at once, and added a smarter cross-tab streaming indicator so switching tabs mid-response doesn't leave you confused about which conversation is still talking. The sign-up page is now a single centered auth panel — fewer steps to a working account — and sidebar collapse and expand animations are smoother throughout.