Anuma connects to your apps — and chat starts showing its work

Connect your apps

Anuma now plugs into the tools you already live in. From the new Connected Apps page you can connect Gmail, Google Drive, and Google Calendar — each independently, so disconnecting one never drops the others — and then decide exactly what each app is allowed to do in a conversation with per-tool allow/deny toggles. Calendar access is scoped to read-only plus event management, nothing broader.

Chat starts showing its work

Replies got more visual. Ask about places and a reply can render an interactive map; ask about a sequence of events and you'll get a vertical timeline; ask for references and image collages appear inline. And on web, Anuma can now generate a working app right in the chat — preview it full-screen, and the app itself can call the model at runtime.

Four new models, and Auto for everyone

Four new open-source models join the picker this week: MiniMax M3, GLM 5.2, Qwen 3.7 Plus, and Kimi K2.7 Code. At the same time, the Auto router is now the default model under every privacy preference — including open-source-only and closed-source-only setups — and its routing always respects the preference you've chosen.

Find anything you've said

Cross-chat search now answers "where did I say that?" by returning cited snippets across your chats, your stored memories, and your artifacts — including history you've imported from ChatGPT, Claude, and Grok. It all runs on-device. Memory recall also got sharper under the hood: in-chat memory now ranks by recency, supersession, and proof-count instead of similarity alone, and consolidation quietly dedupes paraphrased facts — at the same cost.

Council, credits, and polish

Council Mode picked up one-tap defaults — the Council pill opens a send-ready panel of complementary models, and every answer gains an "Ask the council" button. Running out of credits mid-answer is gentler now: Anuma finishes with a short wrap-up and a clear next step instead of cutting off. Subscription management moved into Account settings, and your model-privacy preference is now enforced on every request — on web, on mobile, and on each model in Council Mode.