AI
AI team assistant
Your team asks in a chat app that lets other programs in, or a shared email address, and gets the answer from your own records and playbook.
Updated October 16, 2026
In short
An AI team assistant answers your staff's questions from your own records and written playbook, in a chat app your team uses that lets other programs in, or a shared email address. Each answer says where it came from, and it is built to say so when neither source holds one. The written plan prices it, and the ops retainer keeps it current, with AI usage at cost.
AI team assistant: An AI assistant that answers a team's questions about its own business in plain language, from sources the business chooses: usually the records in one system and a written playbook of how things are done. It is built to look an answer up rather than make one up, so each reply names its source, and a question the sources do not cover gets a plain answer that it does not know, with a pointer to the person who would. At its narrowest it reads and changes nothing.
Key facts
NIST's profile for generative AI, published in July 2024, names confabulation, confidently stated but false content, as one of the risks of generative AI.
Source: NIST AI 600-1, Generative Artificial Intelligence Profile, checked October 15, 2026
NIST's profile for generative AI warns that a model's output can include confabulated citations, and its suggested action MS-2.5-003 is to review and verify the sources and citations in outputs.
Source: NIST AI 600-1, Generative Artificial Intelligence Profile, checked October 15, 2026
Under HIPAA, a covered entity may let a business associate handle protected health information on its behalf if it obtains satisfactory assurance, documented in a written agreement, that the information will be safeguarded.
Source: eCFR, 45 CFR 164.502(e), checked October 15, 2026
What it does with a question
What happens when someone on your team asks, set by the sources you approve in the written plan.
It reads the question
In the chat app or email thread named in your plan, from anyone with the team's access.
It looks it up
In the one system and the one playbook named in your plan, never on the open web.
It answers with its source
In plain words, with the record or the part of the playbook it used, so the person asking can check it.
It hands on what it cannot answer
When neither source holds the answer, it is built to say so and point to the person your playbook names for that kind of question.
What you get
The assistant is only as good as what it reads, so the work starts with your playbook. If it is a shared document already, it goes in as it is; if it lives mostly in people's heads, the written plan says how it gets written down first.
- An assistant in a chat app your team uses that lets other programs in, or a shared email address, named in the plan
- Answers drawn from one system you use and one playbook you give it
- The source named in every answer, so anyone can check it
- One access level for the whole team, set out in the written plan
- A test run on the questions your team asks most, before launch
Where its answers come from
In short
It is built to answer from the sources named in your plan and nowhere else, and every answer shows the source it came from.
A language model can sound sure of itself and still be wrong. Holding it to your own records and your own playbook, and showing where each answer came from, lets the person asking check it in a moment instead of taking it on trust.
It also means a wrong answer has one cause you can fix: a line in the playbook that is out of date, or a record that was never filled in. Fix it once and the answers that read it change with it.
How it goes
We start with the questions your team asks most, gathered from the people who answer them today. The written plan names the system and the playbook it reads, and where your team will ask it, with the price, before anything is built.
We test it against those same questions before anyone relies on it. It goes live in one place first, such as one channel of the chat app named in your plan, and the ops retainer keeps it current as your prices and rules change.
What it costs
Projects start at $2,500. Past that, a team assistant is priced by the sources it answers from and the places your team asks it. More of either is priced in the written plan before anything is built, and so are rules for who may see what.
- Ops retainer
- A monthly fee after launch
- AI usage
- At cost, on accounts in your name
What the smallest version covers: An assistant your team asks in one place it already works, answering from one system's records and one playbook you give it, confirmed in the written plan.
A short call comes first, and the written plan with the exact price follows within two business days, before anything is built. Business days are Sunday to Thursday, and Friday until noon, Los Angeles time; a request after 1 p.m. (noon on Friday) counts from the next business day.
Your rules, your lawyer
In short
We build it and warn you in writing about the legal side. We do not give legal advice, so the rules your AI follows are yours, set with your lawyer.
No legal advice
We flag the questions in writing, and your lawyer answers them. This is general information, not legal advice. Your rules are yours to set with your own lawyer.
A worked example (hypothetical)
Picture a moving company with three people in the office and six crews on the road. All day the crews call the office with questions it has answered many times, like what the company charges to move a piano, or whether a building needs its elevator booked before the truck arrives.
One team assistant would sit in the crews' group chat, on a chat app that lets other programs post in it, and answer from two sources: the job notes in the company's scheduling system and the office playbook, a shared document of prices and rules. A crew chief asks whether Friday's job in Silver Lake has the elevator booked, and the answer comes back from that job's notes, with the job named so the crew chief can check it.
A question the playbook does not cover, such as the price for a pool table, gets a plain answer that it does not know and the name of the dispatcher to ask. The office still sets the prices and books the trucks; the assistant only takes the repeat questions off the phone.
Common questions
Where does it get its answers?
From the system and the playbook named in your plan, not from the web. Each answer says which of the two it used, so an answer with no source behind it stands out.
What happens when it does not know?
It is built to say so plainly and point to the person your playbook names for that topic. A question it keeps missing usually means the playbook needs a line, which the ops retainer can add.
Can it change records, such as moving a job?
Not in the smallest version, which reads and changes nothing. An assistant that updates records needs rules and tests for each change, so it is priced in the written plan as its own work.
Can we keep some answers from part of the team?
In the smallest version everyone shares one access level. Rules for who may see what, such as keeping pay rates from the crews, are built and tested as a job of their own and priced before anything is built.
Can it answer from patient or health records?
No. We do not offer a BAA today, so we do not build it anywhere patient details would reach it, in a record or in a question.
Which chat app can it live in?
One your team already uses that lets other programs post in it, checked for your account in the written plan. Where your team's chat has no way in, the assistant answers at a shared email address instead.
Start with a written plan
Tell us about the business and what you need. We reply with questions, then a written plan that lists the work and its price, and nothing starts until you have read it.