LLM Policy

This LLM Policy covers Avaga AI employees (receptionist, sales specialist, support agent, booking agent, lead manager), Avaga Workforce personalities, Assist Pro on the live line, transcription, summaries, sentiment, and AI CRM logging.

It applies when you run the floor manually with Assist Pro, fully automated after hours, or both. Humans and AI share one workspace brain. There is no side bot with a second inbox.

An AI employee is a seat on your floor with a job description you hired. It can greet, route, take a message, book an empty slot, qualify a lead, or draft a reply. It sits on the same numbers and threads as your team.

It is not a licensed professional. It does not diagnose. It does not invent a legal strategy. It does not guarantee a sale. Library and knowledge-base copy already say this. This policy is that rule in one place.

Assist Pro sits on the live line: next line, transcript, sentiment, CRM fields, follow-up. The useful whispers are short. The dangerous ones are legal, medical, or made up.

A human still hits send when your policy says a human must confirm a price, a prescription refill, or a legal deadline. If the whisper is wrong, the human is still the agent of record.

You hire the personality, train it on your business, put it on a test path, then promote. If you cannot roll back an AI employee in the same afternoon you shipped it, you do not have a studio. You have a prompt in a text box.

Read the transcript the next morning. Fix the line that caused the transfer. Barge-in must work before you leave the seat on a live DID overnight.

A caller who asks for a human must get a human who can see the transcript and the last intent. Handoff without the transcript is a failed automation.

Night shift should leave a transcript, a summary, sentiment, and the next action in the same workspace. If the morning team has to replay a fourteen minute recording to find an address, the AI employee did not finish the job.

Call transcription and voicemail transcription are on the plans. Summaries and CRM notes are records. They follow Privacy, BAA / HIPAA, and your retention settings. Pause before cards. A perfect transcript of a card is a perfect incident.

Do not let an AI employee collect PHI until a BAA and review say it may. Do not let Studio invent a legal strategy. Do not let a lead manager invent a return, a rate, or a licensed recommendation.

If your industry forbids a non-human on the first turn, keep the AI employee on overflow or after hours only, or do not hire that seat.

Where the law or your counsel requires you to say that a caller may be speaking with an automated system, put that line in the greeting and the IVR. Avaga does not hide the roster. You still own the wording.

Avaga uses machine-learning systems to transcribe, summarize, suggest, and run AI employees. We do not publish a rotating model brand on this page, because the floor is the product, not a vendor logo.

Workspace recordings, transcripts, and CRM notes are customer records used to run the floor you bought. If you need a written statement on whether prompts, audio, or tickets are used to train a foundation model, that statement is a sales conversation. Do not treat silence on a marketing page as a never or an always.

Do not prompt an AI employee to bypass pause-and-resume, to ignore a human request, to pretend it has a license, or to scrape a caller for data your form did not ask. Those prompts are Acceptable Use violations even if the model complies.

You configure knowledge, hours, numbers, and promotions. You review transcripts. You decide which seat is live. Avaga provides the workspace. We are not responsible for a script you promoted against this policy.

Product questions: knowledge base and Help Center. Policy questions: hello@avaga.io. Healthcare or legal floors: start with BAA / HIPAA and Talk to sales before you promote.