Test ElevenLabs’ 11 AI assistant with a checked calendar task

Set up the assistant around one small task

Luke’s July 2025 demonstration introduces 11 AI as an ElevenLabs personal voice assistant with calendar and research integrations. The description names Google Calendar, Notion and Perplexity and discusses third-party access through MCP. The free-release language belongs to that recording; it does not establish later account eligibility or connection availability.

1. Open https://try.elevenlabs.io/6fntpnrwyyi0 and follow assistant setup at 1:14. Choose a simple test request whose date, time and intended result you can inspect without relying on private meeting details.
2. Set the voice, tone and behaviour for that task. Keep conversational style separate from the instructions controlling which tools may be used and when a change needs confirmation.

Create an event and inspect its saved details

Google Calendar integration starts at 2:55, followed by a paternity-leave scheduling example at 4:20 and permissions at 5:30. Spoken requests can be understood imperfectly, and correct conversational acknowledgement does not prove a correct calendar record. The saved event is the relevant outcome to check.

1. Connect the intended calendar through the demonstrated route and give one precise test instruction. Include the date, duration and calendar context needed to avoid an ambiguous event.
2. Open the actual calendar entry and compare title, date, start and end times with the request. Correct a mismatch explicitly and inspect the saved result again before trusting a wider scheduling routine.

Combine research with a controlled update

The video adds Perplexity and other integrations at 6:50 and researches and updates events from 8:02. Retrieving information and changing a record are different actions. Keep the evidence behind a recommendation visible before allowing that recommendation to alter the calendar.

1. Ask a narrow research question and inspect the answer and available sources. Identify which returned fact is relevant to the event and which details remain uncertain.
2. Request one specific event update based on the checked information, then reopen the record. Test a follow-up question that does not require a change so you can distinguish conversational research from an instruction to write data.

The takeaway

Judge the voice assistant by the event and evidence it leaves behind. Clear speech, useful research and correct calendar updates are separate capabilities to verify.