> ## Documentation Index
> Fetch the complete documentation index at: https://api.lawtte.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# What our agents are built on

> Lawtte's agents start from intake and follow-up setups proven on live calls at real law firms, practice-area question banks, a library of legal workflows, and guardrails like no legal advice. Your firm's data isn't used to train AI models.

A generic chatbot doesn't know that an injury intake needs the date of the accident,
or that a receptionist must never guess at a settlement. Lawtte's agents start from
the work of real law firms, and the rules that keep that work safe are built in.

## Built from live law firm work

Our receptionist setups were first written and tuned for firms Lawtte answers calls
for every day, across family law, personal injury, lemon law and immigration. Those
live setups are what new receptionists start from.

> [card] 7 practice-area intake playbooks
>
> Family law, personal injury, lemon law, immigration, criminal defense, employment
> law, and a general one. Each has its own intake questions, follow-up questions for
> specific case types, and rules for what's urgent.

> [card] 178 legal workflows
>
> Our skills library covers personal injury, employment, family law, estate
> planning, business law, real estate, litigation, immigration and criminal defense.
> It's public at [lawtte.ai/skills](https://www.lawtte.ai/skills).

> [card] 38 agent templates
>
> Follow-ups lawyers actually send: demand receipts, signed releases, policy
> limits, records requests, RFE documents, mediation prep, signature reminders.

> [card] Your firm's own rules
>
> House rules you write once, like "CC Karina on manufacturer emails", apply to
> every agent the builder designs.

Every intake starts and ends the same way: name and best number first, then the
practice area's questions, then whether they already have a lawyer, how they found
you, and the best time to call back. Where it matters, there's a conflict check on
the other party.

## Guardrails that don't move

These are part of every agent. A prompt or a chat message can't turn them off.

> [accordion] On the phone
>
> * Never gives legal advice or predicts how a case will turn out.
> * Never promises the firm will take the case, quotes fees, or estimates a
>   settlement or a timeline.
> * Treats the call as intake only: no attorney-client relationship is formed.
> * Doesn't collect Social Security or financial account numbers.
> * In an emergency, tells the caller to call 911.
> * Keeps speaking the caller's language.
> * Every outbound call opens by saying it's a virtual assistant calling for your
>   firm, and that the call is recorded.

> [accordion] Before an outbound call
>
> * Calls only people who agreed to be called, and never someone who has their own
>   lawyer on the matter.
> * Calls only between 8am and 9pm in their time zone, and yours.
> * Checks your Do Not Call list, with a daily limit per firm and no repeat calls
>   to the same number within 2 hours.

> [accordion] Texts and emails
>
> * Texts honor STOP right away, go out on weekdays from 9am to 8pm in the
>   person's time zone, and send at most one agent text per person per day.
> * Every email checks your Do Not Email list first.
> * Follow-ups stop the moment someone replies.

## Your data stays yours

* **Not used for training.** The Agent Builder sends AI requests only to providers
  that don't store them or train on them.
* **We don't fine-tune models on your calls.** Your agents get better because you
  tell them what to change: your wording, your house rules, your answers to the
  builder's questions.
* **After every call** you get a summary, the key points, whether they need a
  callback and why, and the caller's tone. Use it to tune what your agent says next
  time.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.