Choose your workspace's AI models, or bring your own

Workspaces now decide which AI models their agents use. A new Models page in settings lists every managed model by provider and lets admins turn each one on or off. Teams that already run an approved model in their own cloud can add it as a custom model, which is the bring-your-own-model path for regulated or cost-sensitive teams. See the current lineup on the models page.

How it works

• Two roles. Agent models appear in the model picker for chat, building, and workflow runs. One utility model handles background work like chat titles and summarizing long conversations, where a fast, cheap model is the right choice.
• Custom models. Any OpenAI-compatible API: a base URL, model name, context window, and capability flags. Keys are encrypted at rest.
• Checked before anything runs. Enabling an agent workflow confirms its model is turned on for the workspace.
• New models. OpenAI's GPT-5.4, mini, and nano joined Claude, Gemini, and MiniMax, and DeepSeek V4 Flash and Pro arrived on April 24 with 1M-token context windows.

Also shipped

• Your default model is saved per member, so it follows you across browsers and devices.