February was the month the AI Assistant grew up. It learned to remember, to see, to show its work — and got a lot more dependable while doing it.

An assistant that remembers

• Assistant memory. The assistant can store information about you across conversations, recall the relevant bits when they matter, list what it has remembered, and forget on request — with per-user privacy kept throughout.
• Rename & regenerate. Rename chats (with automatic titles generated from the content), and regenerate the assistant's reply from any point in the conversation. Empty conversations clean themselves up.

Multimodal and transparent

• Attachments and page images are now sent to multimodal models, so the assistant can reason over what a document actually looks like.
• Visible reasoning time. A collapsed "Thought for N seconds" summary shows when the model is reasoning, with a clearer breakdown of each tool call.
• Chat-context tool. A built-in tool lets assistants list references and search the items attached to the conversation.

Templates, history, and connectors

• Reusable prompt templates with variables, provider testing, and tool selection per template.
• AI code generation inside endpoint and other code editors.
• Generate a C# connector from an existing CSV, with improved reliability.
• Code change history for endpoints, AI tools and scheduled tasks, plus inline documentation with clickable type links.

Reliability everywhere

• Smoother, batched AI chat streaming — with in-flight chunks resent if a connection hiccups.
• Improved file and OneNote ingestion, and re-indexing of legacy macro-enabled .xlsm files.
• A unified save experience (saving / saved / error) and safer code editing that disables save on invalid code.
• A PDF viewer rotate button, facet ordering that floats active filters to the top, and chat titles that match the conversation's language.

---

This overview covers builds v26.2.64177 → v26.3.64833. For the full, build-by-build detail see the changelog.