"You made it! Okay — big question up there, I know.
Stick with me, I explain everything. It's kind of my whole job."
— The Intern, official spokesperson
Five tellings of the same true thing.
The definition — from the founder's board
"XAI Therapy is a therapist-led clinical model that uses explainable artificial
intelligence to illuminate between-session behavior, strengthen therapeutic insight,
and improve outcomes — without replacing human judgment."
Lance Nabers, LPC · 25 years in practice
Your spokesperson
Meet the Intern.
First one to greet you on the island, first one
to admit what he doesn't know. That's why he got the job: a
transparency company needs a spokesperson who can't keep a secret. Ask him anything.
"Okay but what actually IS explainable AI?"
"It means every answer shows its work! Where the data came from, which
clinical framework read it, and who decided. If a tool here can't tell you its
why and its how — we don't ship it. I checked. Twice."
"Is a robot doing my therapy?"
"No! And I'm allowed to be dramatic about this because I am a robot.
Your therapist leads. The AI organizes and cites. A licensed human decides —
every time. We just make the week between sessions visible."
"What happens to my data?"
"It's yours. You can see everything, and you keep your
story — literally, you can take the record with you. A record you can't read or take
isn't care, it's extraction. The boss wrote that one; I just say it louder."
The stance
The data drives the therapy.
The old arrangement points one direction: the modality drives the sessions, and the
sessions drive whatever change survives the week. XAI Therapy reverses the arrow —
present-moment behaviors and cognitions create the
clinical flow. The plan bends to the week the client actually lived.
"We cannot learn or grow if we don't struggle or experience
conflict. There is no learning without a conflict in the mix." — the founder, on why the model never engineers struggle away — only makes it survivable and visible
A therapist sees forty-five minutes of a week that holds 10,080. Everything else —
the sudden spiral, the Tuesday win, the avoided phone call — has always been invisible,
so it was discounted. When the between-session hours produce structured, client-authored,
framework-tagged data, the session stops guessing at the week. The client stops being
the weekly narrator of a fading memory, and becomes the co-author of a living record.
The compass changes hands. Every painting on this site is from our own worlds.
The three principles
What "explainable" is required to mean here
I · Transparency
You can see how it works
No "proprietary" curtain. The client sees what the clinician sees about them. Where AI
helped build the products themselves, we say so — out loud, in the apps.
II · Interpretability
The data means something clinically
A belief rating that dropped. A distortion, named. An avoidance pattern with dates on it.
Readable in seconds, actionable in session — no decoding required.
III · Explainability
Every insight shows its work
Clear, human-understandable reasons for every output — the "why" and the "how."
Kept honestly: evidence + named framework + a human who
decides — never a story pried out of a black box after the fact.
The industry's version of explainable AI asks an opaque model to explain itself afterward —
and the research is clear those explanations are plausible stories, not faithful readouts.
We skip the problem by never building the black box: our tools generate structured,
framework-tagged data by design. See the glass
box, side by side →
The note, and where it is written
Your session never leaves the building.
Most clinical AI sends the hour to somebody else's servers, because that is cheaper to
build. We put the whole thing on the practice's own machine instead — the listening and
the writing both. It is harder. It is the only version of this we were willing to ship.
I · The recording stays
No third party hears the hour
The audio is written to the practice's own disk and the words are made from it there.
It is destroyed the moment the transcript exists, unless you ask to keep it — a recording
of a therapy hour that nothing needs is pure liability.
II · The model never judges
Risk and diagnosis are yours
Asked about suicidal ideation on a transcript where it was raised and explicitly
denied, a model returned nothing at all. "Denied" is a documented assessment; nothing
is a hole in the record. So the model is never asked — your standing lines are placed in
the note afterwards, where nothing it writes can reach them.
III · Every line shows its work
Cited, or it isn't printed
Each field in a drafted note carries the transcript lines that produced it, with
timestamps. A sentence the model cannot point to is dropped,
not printed — you get a visible gap instead of a confident claim the hour
does not support.
Measured on a real session
2 min
to turn a 53-minute hour into words
8,758
words transcribed, on the practice's own hardware
0
of it sent to anyone
And one finding we would rather publish than bury: reading the hour in a single pass, the
scribe reported homework as "not discussed this session." It had been assigned,
forty-three minutes in. A note that turns something documented into a gap is the failure
that matters most here, so the hour is now read in overlapping passes and every field is
checked against the lines that support it. We found that by testing on a real session
rather than a convenient one.
A note still belongs to the clinician who signs it. Where the scribe drafts anything that is
a judgement rather than an observation, it arrives marked unconfirmed
and the note cannot be signed until a human has read it and accepted it. That refusal
lives in the software, not in a policy document.
And there is a consequence worth saying plainly. Every other clinical AI
asks you to sign a Business Associate Agreement, because your clients’ words are going
to sit on their servers and somebody has to be liable for that. Here, nobody holds your
record but you — so there is no agreement to sign with us, and no third party whose
breach could become yours. That is not a policy we are promising to honour. It is an absence
of anyone to promise anything.
Consent that actually informs: what is captured, why, and who sees it — the client,
always. The story-world begins, and the first task is the smallest possible yes.
02
Therapeutic rapport
The relational floor — unconditional positive regard, in person and in code. No shame in
the data, ever. The only assignment is to return.
03
Assessment — continuous, not episodic
Baseline the person, not the intake hour. Two weeks of lived data beat any single
interview — and in this model, assessment never closes.
04
Behavioral model learning
The namesake phase: the client learns the system that runs them — trigger, thought,
behavior, consequence — and becomes explainable to themselves.
05
Task completion
The between-session act is the unit of change. One task at a time, honestly verified,
reinforced immediately. Avoidance has a timestamp now.
06
Skill building & generalization
From tool-use to trait. Skills re-run in harder contexts, taught back to others,
until they leave the app and enter the life.
07
Integration & ownership
The client reads their own arc against their own baseline — then keeps it. The record
is theirs to hold and take. That is the point.
AI is already in the therapy room. Notes, scribes, chatbots — most of it
black-box. The only question left is whether it can explain itself to the person whose life
it touches.
Trust is the treatment. The alliance predicts outcomes across modalities.
An inexplicable system inside the alliance corrodes the active ingredient itself.
Clinicians are accountable; black boxes are not. A licensing board, a
court, a client — all may ask why. "The model said so" is not an answer a
professional can stand on.
Clients own their stories. A record the client can't read or take with
them is extraction, not care.
The between-session hours are where the life happens. The only honest way
to see them is data the client authored, in the open, on purpose.
Some help doesn't fix. It holds.
The worlds — where the model lives
Engagement is the delivery system
Hand someone ninety-seven tools and they try two.
Wrap them in a world worth returning to, and they find the one they needed. Every painting,
every robot, every island below is ours.
LANCExai — the flagship
Five acts, thirty-one challenges, real clinical instruments (CBT, DBT, EMDR pacing,
parts work, grounding) — one step at a time, on an island that keeps you.
The family crossing
Couples and families work the model together — pass-the-phone firsts where every
hand that helped is named. The island only yields to together.
The Lighthouse — the clinician's side
Where the glass-box data becomes a reviewed, explainable, exportable insight — and every
AI reply carries a visible receipt: which layer answered, which rules fired, how.
Lance Nabers is a Licensed Professional Counselor, twenty-five years in practice. He began
as a child therapist who used robots to help kids rehearse new behaviors — "the robot
does this; what do you do?" Twenty years later the robots are back in the room,
opening the work up instead of hiding it. That is the honest origin of Wayward Robots,
and of everything on this page.
For clinicians
Be one of the first to practice in the open.
The founding-clinician list is small
on purpose: real therapists, early access, and a direct line to the founder. Tell us where
we're wrong — that's the job. ("I read every reply.
Loudly." — the Intern)
