Before you try it

A general chatbot plays a client the way it guesses one might be: too articulate, quick to open up, happy to drop the role when asked, and with no memory of last week. Untangle is built for training:

• Real cases, not guesses. Each client has a full history, personality, memories and triggers, and things they won’t give up easily.
• They behave like clients. Trust is earned slowly, a clumsy question closes them down, and a rupture needs repairing. They never step out of the role.
• A story over several sessions. They remember what happened last time, and how you work changes how their story ends.
• Feedback a chatbot can’t give. We know what was really going on with each client, so the debrief can compare your formulation with it and point to the exact moments that helped or hurt.
• Progress you can measure. Skills tracked session by session, for you and for your supervisor.

Both the big picture and the moment-to-moment.

• Macro-skills: case formulation (your notes compared with what was really going on), building and keeping the therapeutic alliance, noticing and repairing ruptures, pacing (when to push and when to wait), and keeping a session focused.
• Micro-skills: reflecting and validating, open and closed questions (and long runs of them), well-timed silences, what you do as well as what you say, and how each response landed with the client.
• Depending on the client: risk assessment, holding boundaries, working with a client who doesn’t want to be there, and following up homework.

You can go back to any moment, try a different response and see how the client takes it.

No. Each client is built from patterns in real therapy, with every identifying detail changed and their story rewritten. None of them is a real person.

Claude, an AI model made by Anthropic, guided by each client’s case.

Clients show as much as they tell: a shrug, a long pause, a glance away. Trust builds slowly and can be lost with a clumsy question, and repaired. Each client remembers earlier sessions.

Yes. Search for a presenting problem, a scenario or a skill on the home page. If we don’t have it yet, we note it: that’s how we decide which clients to build next.

With a pilot, yes: each trainee’s sessions, and their skills and habits session by session.

No. It gives trainees practice before and between real clients, and gives supervisors more to work with: what actually happened in a session, not only what the trainee remembers.

Practising gives feedback after every session. Doing it for real holds the feedback until the client’s story ends, so you see where your skills really are. In a programme, these are the sessions your supervisor sees.

Up to an hour, like a real one, with a reminder before the end. You can end a session at any time.

You can try it free. For training programmes we start with a free pilot, then agree a price per trainee. Plans for individual practitioners are coming.

In Sydney, Australia. Session text is sent to Anthropic in the United States to play the client and write feedback, and isn’t used to train its models. The privacy page has the details.

A web browser. It works on a phone, but a computer or tablet is more comfortable for a full session.