Choose an AI model
Use Recommended by default or choose another approved model for an AI conversation or Meeting Minutes summary.
Integrity shows only models approved for the task you are performing. You can keep Recommended for the default experience or deliberately choose another available model from the AI model menu. Each choice shows its provider mark, model name, cost, and a short description of what the model is suited for.
The lists for Integrity AI conversations and Meeting Minutes can differ. A model appears only when it is approved for the complete workload, including the required output shape and safety boundaries.
Integrity AI conversations
Your choice belongs to the current conversation. It remains selected for later messages in that conversation until you change it. Each assistant response keeps the model label that was used, so older responses remain understandable if the current selection or available list changes.
Meeting Minutes summaries
Open an editable Meeting Minute and select Process notes or Process Meeting. In the confirmation dialog, choose a model from the AI model menu, then confirm processing. The choice applies only to that request. If you reload the page before processing, the dialog returns to the current Recommended model.
An existing result shows Generated with in the AI Summary card. Choosing a model in the processing dialog does not change the label stored with an older summary. Each successful regeneration becomes the latest result without deleting earlier successful results; a failed regeneration leaves the previous latest result intact.
When history exists, Previous summaries (N) loads one older generation into the existing summary card. Its AI Summary and Recommended actions always come from the same model run and are labeled with model, time, source, and whether the generation matches the current Notes. Historical actions are view-only; use Back to latest to return to actionable current output.
Generated summaries, decisions, Actions, and open questions remain reviewable AI output. Choosing a model does not overwrite human Notes, finalize the Meeting, or create Issues automatically.
Different models can produce legitimately different emphasis or detail in a summary. Review the generated result against the Notes or transcript before applying or acting on it.
If a model is unavailable
Integrity never silently substitutes another model. If the selected entry was removed, changed, disabled, or is temporarily unavailable, choose an available model and submit a new request. A chat model can also disappear when its tested AI Workspace compatibility expires or no longer matches its exact configured version. Refreshing the page checks the current qualified list. If no chat model is qualified, the composer remains unavailable until an operator restores a passing entry; Integrity does not choose a different model automatically. Work that was already generated keeps its stored model label.
AI Workspace compatibility and Meeting Minutes approval are evaluated separately. A model becoming unavailable for chat does not by itself change whether it can process a Meeting Minute, and the reverse is also true.
Your model choice does not change workspace permissions, source access, available tools, review authority, or restricted-content behavior.
External model providers
Some approved choices may be operated by an external model provider. Integrity still applies the same permissions, bounded context, tool, citation, and review rules, but the authorized prompt or Meeting source must be sent to that provider to generate the result. Your organization controls which qualified entries are available. If external-provider use is not approved, the menu remains fully usable with the managed Recommended model only.