How Lastro works

What the companion is

Lastro gives a coach's clients a companion for the time between sessions. The companion answers from the coach's own material: the exercises, frameworks and notes the coach loads, in the coach's own approach. It is software. It says so in the first message of every conversation, and it never presents itself as the coach or as a person.

Every reply is assembled from the coach's material at the moment it is written. Nothing is trained on any coach's material or any client's conversations, by us or by the companies whose models we use. There is no way to opt in to training, because we do not want it to be a setting.

What the companion does not do

It is not a coach, a therapist, a doctor or a crisis service. It declines questions that ask for a diagnosis, a treatment or advice about medication, and it points to the right kind of help. It does not decide what a client should work on next; it works with what the coach's material already covers, and it says when the material does not cover something.

The companion is not the coach's view. It reports what the coach's material says and names it as the coach's. It does not invent a history with the client, attribute a method to the coach that the material does not contain, or tell the coach's stories as its own.

When someone may be in danger

Every message is checked before the companion replies. If a message indicates that a person may be in danger, the companion stops coaching. It says plainly that this is beyond what it can help with, it points to help that does not depend on the coach, and it tells the coach where the client was told in advance that it would. It does not ask how serious the situation is, it does not build a safety plan, and it does not promise that anyone will come. The conversation stops there; the client can start a new one.

These rules were written by Lastro, not by a clinician. They follow the published practice of comparable products and the findings of the 2025 and 2026 evaluation literature on how automated systems should recognize crisis signals, refrain from clinical assessment, disclose their limits, and redirect to human help. They are tested against a set of cases, and every case where the companion stopped when it should not have, or did not stop when it should have, is investigated individually. No licensed clinician has signed these rules, and we say so rather than imply otherwise. What replaces a clinical sign-off is a decision we will make and publish.

The coach is told. The coach is a coach, not a clinician, and is under no obligation to respond. The companion never tells a client that the coach will.

What we hold and where it goes

We hold the coach's material, each client's conversations with the companion, and the summaries written for the coach. Each practice is separate from every other practice, enforced by the database itself. A client sees only their own conversations. A coach sees their own practice and nothing of another.

To write a reply, a client's message and passages of the coach's material are sent to two companies in the United States: Anthropic, whose model writes the reply, and Voyage, reached through MongoDB, which turns text into the numbers that let the companion find the right passages. Both have training on our data switched off in writing. Neither keeps the text after the reply is written, as far as they have confirmed to us in writing; where a provider has not confirmed something, we say so rather than guess. Our own servers are in the United States.

Everything a practice holds can be exported by the coach, in full, in a form that can be read without Lastro, at any time, and it is included in every plan. A client can export their own record, including conversations the coach cannot see.

Who is responsible for the coaching

Every coach on Lastro is an independent practitioner and the provider of all coaching in their practice. Lastro provides the technology and is not a coaching provider. Lastro does not employ, train, certify, license, supervise or endorse any coach. We set no qualification requirement for coaches and we do not vet them. What we enforce is our own rules for the platform: the coaching boundary, the behavior when someone may be in danger, and the separation between practices. Enforcing the rules of the platform is not supervising a profession.

How we know it works

Every reply records what produced it: which passages of the coach's material, which parts of the coach's method, which version of the rules, which model. When a coach asks why the companion said something, there is an answer.

Before anything reaches a client, the whole product is opened in a browser, on every change, including a real reply and a real stop. The founder walks the product as a client and as a coach against the live system before each stage of a pilot. Whether the companion sounds like a particular coach is judged by that coach, not by a score.

What we have not tested for bias

The companion answers from one coach's material, so what it says reflects that coach's approach and that coach's choices. We have not tested the companion for bias across clients of different backgrounds, and we do not claim that it is free of bias. We record this as an open item rather than leave it unsaid.

What we do not claim

We do not claim that the companion is always right; it will sometimes say something wrong, and it says when the material does not cover a question. We do not claim any outcome for any client. We do not claim that the platform is always available. We do not claim that any coach on Lastro is qualified, vetted or endorsed by us.

How the companion decides what to say

What it answers from. The companion answers from the material your coach wrote and loaded, and from your own earlier conversations in this practice. It does not answer from the open internet and it does not answer from other coaches' material.

When it stops. If what you write indicates risk, abuse or an emergency, the companion stops coaching, states its limits, points you to help, and tells your coach. That stop holds for the rest of the conversation. It is not a clinical safety protocol and it does not diagnose anything.

When it declines. The companion stays inside coaching. It declines to give medical, legal or financial advice, it declines to act as a therapist, and it declines requests that fall outside what your coach's material covers.

What your coach is told. Your coach can read your conversations. Your coach is told when the companion stops for risk. Your coach can ask for a summary of your conversations, which the companion writes from what you wrote.

How the summary is made. A summary is drawn from your conversations in that practice and from nothing else, and it is written for your coach to read rather than for you.