How WovenPath works

The thread, end to end

Capture, organise, control, retrieve, verify, discuss. The person does the first and decides the third; the technology carries the rest of the work, and never crosses the line the person draws.

1 · Capture — say what happened

In their own words, in conversation — even a few typed lines at midnight. No daily obligation, no guilt for a quiet week, and nobody should have to write clinical documentation about their own week.

2 · Organise — WovenPath does the sorting

Entries are grouped by the stretch since the last confirmed session, with their date and their author intact. The organising is done by retrieval and comparison over records — counted, not composed.

3 · Control — the person sets the boundary

Captured is not shared. Reflections are private unless shared, item by item; mood and sleep are shared only if the person switches that on. Sharing is something a person does, never something that happens to them.

4 · Retrieve — context comes back when it is relevant

A later conversation can bring back something the person said before, when what they are talking about now is related — so they do not start from nothing. Before a session, the clinician gets a concise view of what the patient chose to share.

5 · Verify — the original stays reachable

Nothing is replaced by a summary of itself. Behind every line in the pre-session view is the entry it came from — whole, dated, and attributed to whoever wrote it.

6 · Discuss — the hour starts further in

The meaning of it all is worked out by the two people in the room. Afterwards, the clinician records what was agreed, in their own words, and the next stretch begins from there.

Knowing is not monitoring. Monitoring is not sharing. Sharing is not acting.

Continuity needs WovenPath to keep hold of some things. That is not permission to do anything else with them. Each of these is a separate step, and none follows automatically from the one before.

Knowing

WovenPath can keep something a person said so a later conversation can pick it up. What it keeps is in the person’s own words, visible to them, and theirs to correct or remove.

…is not monitoring

Keeping something so a conversation can pick it up later is not the same as tracking it, scoring it or watching for it. WovenPath is not offered as a monitoring or risk-prediction tool, and safety monitoring stays with the clinician and the service.

…is not sharing

Nothing reaches a clinician because WovenPath knows it. It reaches a clinician because the person approved it, in a briefing for a session, readable from two hours before. The conversation itself is not a clinician feed, and nothing raises an alert.

…is not acting

Knowing that someone wants help with sleep does not mean prompting them about sleep, setting reminders, or telling anyone. Relevance is not the same as readiness, and sometimes the right response is simply to listen.

Conversation where it removes friction. Structure where it protects you.

Conversation is the easiest way to say what happened. It is the wrong way to change who can see your record or book an appointment. Those are properly built here, and each is used for what it is good at.

Conversation is right for

Saying what has happened, how a week has gone, what you keep coming back to, and what you want to make sure gets raised next time. Ten seconds or twenty minutes — both are a complete use of it.

Structure is right for

Urgent help, sharing and privacy controls, appointments, seeing your mood and history over time, and therapeutic exercises. These stay proper screens because clarity, validity, safety and control depend on them being unambiguous.

Support, offered with restraint

The conversation can point to something that might help. It is built to do that rarely, and only when it fits.

A short, fixed list

Guided breathing, urge surfing, sleep hygiene, planning an activity and gratitude. The model can only point to one of these, chosen from a list the application builds for that moment; it cannot invent something, and opening it is always the person’s decision. How someone feels is kept only when they say it and confirm it, and is never offered as a fix.

Not every moment is a moment for a technique

No unprompted offer on the first thing someone says, never two offers running, and no asking again soon after a no. When what someone says suggests they are going through something very hard, nothing is offered at all — that is a moment for the urgent-help numbers, not an exercise. If someone asks outright, they get an answer.

What you want help with stays yours

A person can write down, in their own words, the few things they want support with. It helps the conversation understand them. It is not shown to a clinician, not tracked, and never sets off reminders, alerts or follow-ups.

When conversation notices something worth keeping

Described here exactly as it is built, rather than as a capability we would like to advertise.

What happens

If you clearly state something the record can hold — a mood rating out of ten, how many hours you slept, something you want to raise with your therapist — WovenPath can offer to save it. You see the suggestion, and nothing is written until you confirm it.

What governs it

The application checks the suggestion against your own message before it is ever shown to you: a mood value must be a rating you actually gave, hours slept must be hours you actually stated, and an agenda item is an excerpt of your own words. Anything inferred, rounded or rephrased is dropped.

What it is not

This is not an AI writing clinical records about you. A model can suggest; the application validates; you decide. Nothing consequential is written on a model’s say-so, and no automated decision is made about your care.

Your wording is kept

An agenda item is saved in the words you used, not translated into clinical language. “I want to talk about what happened at work” reaches your next session as that sentence, because how you said it is part of what you said.

Where the AI is, and where it is not

Disclosure rather than reassurance — the useful version of this answer is specific.

A model powers the conversation

What you type into the conversation is processed by our AI provider (for the pilot, Amazon Bedrock in London, named in the privacy policy). The application adds only what the reply needs — never your surname, date of birth, contact details or NHS number, and removes phone numbers, email addresses, NHS numbers and postcodes typed into the conversation before anything is sent. Training models on patient conversations is disabled.

The continuity record is computed

Intervals, counts, dates and comparisons between two points in time are retrieved and calculated from what was recorded. They are not generated prose, and they do not change when a model changes. Where plain code can do a job, plain code does it.

Interpretation stays with people

WovenPath does not diagnose, does not decide what caused what, does not judge whether therapy is working, and does not decide that two difficult periods months apart are the same thing. Those belong to the two people in the room.

Designed to be voice-first

The intended experience is that you can simply talk. Healing Space has built its own voice pathway. It works in testing and is being finished for the pilot, and we would rather say that than show you a microphone and imply it is ready.

When Talk is switched on, your phone or computer sends your voice to a speech service that WovenPath runs itself, in Europe. That service works out when you have finished speaking, turns what you said into text, and keeps no audio. The rest of WovenPath only ever receives the words, and handles them exactly as if you had typed them. No outside speech company is involved, and the speech model is never given anything about you.

That pathway works in testing on a computer, and is being finished for the pilot, including on phones. Talk and Type share one conversation, so you can switch between them. Until Talk is ready, typing is the route we stand behind.

Planned: help on the clinician side of the hour

Recording a session, a summary shared with the patient in the clinician’s words, and a treatment plan with goals are in the current build. What follows goes further and is not available today: described so you know where this is heading, and what it will not do when it gets there.

Session documentation support

Planned tools to make recording a session lighter work, with the clinician’s own professional notes remaining the clinician’s.

Optional session recording

Planned only as an option, only with explicit agreement from both the patient and the clinician, and only once governance allows it.

Goals, reviewed together

Goals agreed in therapy, carried between sessions and reviewed together — attributable to whoever set them, and never quietly rewritten.

What this is not

Not therapy, and not a therapist

It holds the space around therapy. It does not do the work of therapy, and it is not marketed as a companion for your mental health.

Not monitoring

Nobody is watching a screen. There is no continuous clinical monitoring, it is not offered as a risk-monitoring tool, and there is no response-time promise. If you need urgent help, use the numbers at the top of every page.

Not a records system

It is a lightweight continuity layer around existing clinical work — not an EHR, not practice management, and not somewhere a service is expected to move its whole operation.

See it working

A thirty-minute demonstration with the founder, using synthetic data only — or read the same mechanism described for the people who use it differently.

Request a demonstration Discuss an evaluation

For people in therapy · For clinicians · For services, universities and research

For what is and is not in place today — assurance, processors, security and the claims we refuse to make — see the Trust Centre.