Skip to main content

June 8, 2026

Highlights​

This release is a big step forward for how chat feels day to day, where you can reach Devs.ai, and how admins control models:

  • A rebuilt chat experience — chat now runs on a new conversation engine built for longer, tool-heavy threads. Responses stream more smoothly, interrupted chats recover more cleanly, and in-progress answers are more likely to survive a refresh or brief connection drop. This foundation also sets the stage for more advanced chat capabilities arriving over the next few weeks.
  • Microsoft Teams integration (preview) — organizations can connect Devs.ai to Microsoft Teams; available by request while in preview.
  • LLM provider status — a dedicated status page and in-app overlay show real-time availability for the providers your organization uses, scoped to your enabled models.
  • Per-surface model configuration — admins can control which AI models are available on chat, agents, routines, and App Builder through a new settings drawer; pickers and defaults respect those boundaries.

We also shipped a week of MCP admin UX improvements across the June 3–9 release window.

New features​

A rebuilt chat experience​

Chat is now powered by a new conversation engine rolled out this week. The chat you already know is still there — same agents, same tools, same workflows — but the foundation underneath is built to keep up with longer conversations and more complex back-and-forth.

What you should notice:

  • Smoother answers as they arrive — streaming responses feel steadier, with formatting that holds together better while the AI is still typing
  • Less lost work — if you stop a response, approve a tool step, or need to step away mid-answer, the chat is more likely to land in a sensible state instead of leaving you with a broken thread
  • Better recovery after interruptions — refreshing the page or losing connection briefly while the AI is still responding is less likely to force you to start the answer over from scratch
  • More room for complex chats — conversations that lean on tools, research, images, and follow-up actions are handled more consistently from first message to last

This is deliberate groundwork, not the finish line. The new engine is what makes the next wave possible — more capable agent collaboration, richer workflows, and chat experiences that go well beyond a single back-and-forth. Those pieces start landing over the next few weeks. Watch this space.

Microsoft Teams integration (preview)​

Devs.ai for Microsoft Teams is in preview and available by request. Reach out to your account team to enable it for your organization.

Once enabled, organizations can connect Devs.ai to Microsoft Teams:

  • Install from org settings — start the Teams app installation flow from the Integrations page
  • Tenant-aware routing — installation endpoints resolve the correct organization by tenant ID

LLM provider status​

A new provider-status experience helps you see whether upstream LLM providers are healthy before you start a long chat or routine:

  • Status page — view availability for Azure, Google Cloud / Vertex, and other Statuspage-backed providers
  • In-app overlay — a banner surfaces active incidents or degraded providers without leaving your current page
  • Org-scoped — status is filtered to providers and models your organization actually uses, and gated behind a dedicated permission

Status is refreshed on a schedule with caching so the UI stays responsive while staying current.

Per-surface model configuration​

Admins can now control model availability per surface instead of relying on a single global list:

  • Model config drawer — enable or disable models independently for chat, agents, routines, and App Builder
  • Surface-aware pickers — model dropdowns only show models allowed on the surface you are working in

Routines in org settings​

Routines now appear in organization settings with clear agent precedence rules, making it easier for admins to see how scheduled agents are configured at the org level.

Improvements​

MCP administration​

Connector and tool-template configuration now opens in wider side drawers instead of cramped modals, giving admins more room to review authentication, settings, and allowed tools before saving.

Models​

  • Gemini 3.5 Flash tool calling — Gemini 3.5 Flash now supports tool calling
  • Image generation on Auto — chats using the Auto model can now attach the image generation tool when appropriate

Model updates​

  • Claude Opus 4.1 is scheduled for deprecation in August 2026

Bug fixes​

Various fixes to chat reliability, routines, provider status, and document handling.

⌘J