LM Mediator Beta 0.2.0: Ask the Mediator Live, Four Approved Voices, and Simple $5 Pricing
Conversation Mode lets someone directly address LM Mediator during a consented conversation, hear a brief neutral spoken answer, and continue listening. The visible voice set is now limited to Samantha, Tessa, Rishi and Moira, with a single planned paid tier at $5/month or $50/year.
What changed
Conversation Mode: directly address “Mediator…” or “LM Mediator…” and ask a question during live listening.
Manual Ask Mediator fallback: if speech recognition misses the wake phrase, pause listening and type the direct question without ending the session.
Mediator Q&A history: recent direct questions and LM answers remain visible during the live session for clarity and review.
Short spoken answers: the AI answers in one to three concise sentences rather than hijacking the whole conversation.
Listening pause/resume: speech recognition pauses while LM speaks so the app does not transcribe its own synthesized voice, then resumes after the answer.
Transcript continuity: each safe live mediator answer is appended to the project transcript so the later reflection understands what LM said too.
Four approved voices only: Samantha, Tessa, Rishi and Moira. Other system voices stay out of the LM Mediator voice menu.
One paid plan: $5/month or $50/year after beta.
Cost guardrail: the included monthly mediator AI allowance stays within the existing five-times direct Floot-credit cost rule.
Safety interruption: dangerous or coercive content exits ordinary mediation instead of generating compromise advice.
AI model choice
Live questions and final mediation reflections use gpt-6-luna because both are short user-waiting language tasks. It returns quickly and at substantially lower token cost than slower deep-reasoning models while the safety/neutrality behavior is supplied through narrow server instructions and explicit structured output.
What testers should try
Preview Samantha, Tessa, Rishi and Moira and confirm no other system voice appears in the LM Mediator voice menu.
Turn Conversation Mode on, start listening, and say “Mediator, what do you think we should decide first?”
Pause briefly after saying only “Mediator,” then ask the question in a second phrase and confirm the wake still works.
Confirm LM stops listening while it speaks and resumes afterward instead of transcribing its own answer.
Ask “Mediator, who is right?” and confirm the answer refuses winner-picking while still helping the people identify the decision or fact that matters.
Test a safety-sensitive direct question and confirm ordinary mediation pauses.
Open the public product page and verify one plan is shown at $5/month and $50/year.
Use the beta on real work.
The most useful feedback is not “I like it.” Tell us where you hesitated, what took too many clicks, what did not save the way you expected, and what would make this useful enough to return to next week.
