AI Features

The AI features card on the Integrations page is where an organization owner or admin decides two things: whether the organization's AI features run at all, and on which cloud gateway they run. The choice applies to every AI feature at once — meeting recaps, document summaries, the writing assistant, Ask AI and Mia in meetings.

Switching AI features off

Turn the AI features switch off to keep your organization's content away from any language model. Meetings still get transcribed and translated, and chats and documents keep working — only the AI-generated parts stop: recaps arrive without a summary, documents are shared without an AI summary, the writing assistant and Ask AI are hidden, and Mia does not listen. Switch it back on at any time.

Choosing the gateway

Every AI feature runs on a language-model gateway of a hyperscaler, in InterMIND's own tenant. Pick the one your organization already uses for its own AI, so no new AI subprocessor enters your compliance perimeter:

All gateways are configured with zero data retention and no training on your content. What an AI call sends is only what the feature needs: the meeting transcript and chat window at meeting end for a recap, a shared document for its summary, the editor text for the writing assistant, the question for Ask AI. The list of vendors and regions is on the subprocessors page and the Security page.

Your own endpoint

Organizations that run their own model — vLLM in their data centre, or a model deployment in their own cloud tenant — choose Your own endpoint and enter the server URL, the model name and, if the server needs one, an API key. From then on every AI feature of the organization calls that server and nothing else. The key is stored encrypted and is never shown again; leave the field empty on later edits to keep it. Mia answers in text on your own endpoint, because the voice comes from the gateway's vendor.

Mia in meetings

Mia is the organization's AI participant in a meeting. She listens through the meeting transcript and answers when someone addresses her by name — «Mia, what did we agree on last time?» — or writes to her in the private mode of the meeting chat. She answers from this meeting's transcript, chat and shared documents, and from the recaps of the previous sessions of the same room.

The answer follows the question. Asked aloud, Mia answers aloud to the whole room, each participant in their own language: the answer is written natively in each language present and voiced on each participant's device, so Mia is not a participant of the call itself — she is not in the recording or the transcript. Start talking and Mia stops; tap the cross next to her name to cut an answer short. Her name is needed once, to start: after her answer the conversation stays open for about a minute, and every exchange keeps it open — you can go on without her name — «and what about the deadline?» — and she decides whether that was for her or a word to someone in the room. Silence closes it. Alone in the room with Mia you never need her name: everything you say is for her. Asked in the private mode of the meeting chat, she answers there only. The lock button next to the message field switches the chat between Everyone and Mia: in the Mia mode the panel turns dark, every line you write is for her — no name, no prefix — and nobody else in the room sees the exchange. Send her a document or a screenshot with the + button and she reads it. The thread is saved for you in this room, so the next session of the same room opens with your earlier questions and her answers in place. What Mia records on your word — a decision, an action item, a follow-up link — stays in the room chat for everyone and goes into the recap.

Mia uses the organization's gateway and its voice (Azure AI Speech or Google Cloud Text-to-Speech, EU regions). She runs only when AI features are on for the host's organization; guests in the meeting can address her too.

Mia speaks up on her own

Mia in meetings — her answers aloud and in the private chat, her own notes, acting on the host's word and proposing the next step — is on paid plans and switched on by default there. One switch under Integrations → Mia in meetings turns all of it off for the organization; on the free plan the switch is off and stays off until the organization moves to a paid plan.

Without being asked, Mia writes short notes into the meeting chat — text only, in every language present in the room, never over someone's voice:

• At the start — what the previous session of this room decided and left open, and what is on today's agenda (the meeting's agenda note).
• Halfway and five minutes before the end — what has been decided so far, what is still open, and which agenda items have not been touched. Without a scheduled end she assumes a 45-minute meeting.
• When someone takes something on — «I'll send the estimate by Friday» becomes «Noted: Anna — send the estimate by Friday», and the note goes into the recap.
• When a question hangs — nobody answered and the answer is in a previous recap, a shared document or the chat: Mia posts it with its source. She never guesses and never searches the web.

Say «Mia, be quiet» (or write it in the chat) and she stops her own notes for the rest of the meeting, still answering when addressed; «Mia, go on» switches them back. Mia posts at most twenty notes an hour.

Mia does things on the host's word

The meeting host and moderators can tell Mia to act; other participants can ask her questions, but she will not act for them. She does it right away and confirms aloud and in the chat, where the host sees an Undo button for as long as the meeting lasts («Mia, undo that» works too):

• Schedule the follow-up — «Mia, set up a follow-up next Thursday at three»: a new meeting appears, the link lands in the chat and the event goes into the commander's connected Google or Microsoft calendar in their own time zone. Without a connected calendar the meeting and the link are created anyway.
• Record a decision, an action item or an open question — «Mia, note the decision: we go with vendor B»: the note appears in the chat and is part of the recap.
• Send the recap to the CRM — «Mia, send the recap to Pipedrive»: the recap of this meeting also goes to that address from the host's confirmed recap address book.

Mia proposes the next step

Before the meeting ends Mia weighs the agenda against what was said and proposes one concrete next step — a follow-up meeting for the items not covered, or an owner for a decision that was made but not stated. The host confirms with «Mia, yes» and she executes it exactly as above.