Team model connections
Set up and verify a team model connection, select it for Mind and Study requests, and understand revisions, credentials and current limits.
Availability and scope
Where enabled for a Team workspace, team admins can add a public HTTPS model endpoint. Supported protocols are OpenAI Chat Completions, OpenAI Responses and Anthropic Messages. A connection changes the selected answer-generation path; supporting analysis still uses Minds models.
Add and verify
Open Team settings and add a model connection. Enter a label, endpoint URL, model identifier and API key, then save. Run verification before selecting it. Verification sends fixed synthetic requests to check text, structured output, streaming and tool calls. These requests can incur provider charges; a failed capability remains unavailable.
Select and retry
Manage provider preferences and model connections in Settings → Team → AI configuration. Study and Mind chat inputs do not display model selectors. For an explicit custom connection in API or MCP requests, pass its verified ID and current revision as modelConnection. Studies require text and structured output; Mind chat requires text, streaming and tools. Queued questions and retries retain the selected connection revision. Changing settings invalidates verification. Reverify before using the new revision; an old queued revision does not silently switch to it or to the default model.
Credentials and limits
Credentials are encrypted for the team and are not returned in connection lists. Revocation removes the stored credential and prevents further authorized use. Provider charges are separate from Minds plan usage. Images, video and visual website inputs are not supported by connections in this release. Use a request without a custom model connection for visual inputs. Text documents can use their existing analysis path. Adding a connection does not authorize dataset export or training.
Troubleshooting
If the connection is unavailable, check your Team access, endpoint, model identifier and current verification. A timeout, revoked key or unsupported capability returns a failure without an implicit model fallback. Do not repeatedly retry an unchanged failed setup.
