Everyone hears "AI" and pictures a chatbot. We built something else.
Most weeks, a practitioner asks us where AI belongs in parts work — and where it doesn't. It's the right question to ask of any tool near the therapy room, and it deserves a plain answer. This page is ours: who we are, why we built Parts Companion, and the principles that decide what we will and won't build.
A philanthropic lab, not a startup.
Parts Companion is built by Evolve Labs, a division of Evolve, a philanthropic organization whose charter is the evolution of consciousness. We think about that charter through two pillars: the alleviation of inner suffering and the facilitation of personal transformation.
The Evolve Foundation has supported that mission for years, including its long support of research in what is now called the psychedelic renaissance. Evolve Ventures backs mission-aligned entrepreneurs and projects like Insight Timer at the earliest stages, back before a meditation app on your phone was an ordinary thing.
Evolve Labs exists for a different situation: when something we believe should exist in the world is not being built, usually for market reasons. There is not enough money in the thing. Venture capital isn't interested, and the market doesn't solve it. So we provide philanthropic support and build it ourselves.
Funds the research
Years of philanthropic support for inner work, including long support of the research behind the psychedelic renaissance.
Backs the builders
Early backing for mission-aligned entrepreneurs, from before a meditation app on your phone was an ordinary thing.
Builds what the market won't
When something should exist and nobody is building it, we provide philanthropic support and build it ourselves. Parts Companion is that project.
A companion for the work, not a stand-in for the person.
When Labs went looking for its first project, we surveyed what we call inner technologies: parts work, the Enneagram, Nonviolent Communication, meditation, breathwork — the tools humans have refined over generations for working with the inside of a life. Parts work won out. Many of us had met it in our own personal work and felt its unusual potential to reduce suffering and facilitate transformation.
There were already plenty of AI therapy chatbots in the world, products built to stand in for the practitioner. What did not exist, and what nobody seemed to be building, was a companion for the practitioner and client working together. We honor the sacredness of the therapy container and the human relationship at the center of it, and we believe most of what this technology can offer therapeutic work has nothing to do with replacing the person across from you.
That conviction became Parts Companion. After each session it produces artifacts that serve both sides of the room: session notes, a map of the client's parts in the client's own words, meditations personalized from the session itself, consultation feedback that helps the practitioner keep growing. The therapist remains the therapist. We built it with more than twenty IFS lead trainers, certified therapists, and expert clinicians — you can meet the full clinical team here.
"A therapist once asked me, in a public forum, to promise that we would never build a therapy chatbot. I could answer honestly: we have had the clinical team, the technology, and the resources to build one for years. We haven't, because it runs against the impact we're here for — and I see nothing today that would change that."
One arc keeps reappearing.
Across inner technologies, the same shape shows up again and again. Psychedelic-assisted therapy formalized it as preparation, journey, and integration. Anthropologists described it in rites of passage over a century ago: separation, threshold, return. In plainer words: awareness, experience, integration.
The most useful thing we can say about AI in this field is that it serves the two ends of that arc — and the middle is where we tread lightly.
Awareness
A clearer understanding of one's own system: the parts that are present, the patterns that repeat, and the ways those patterns shape experience and relationships. A client who can see their own system — parts as they actually named them, patterns surfacing across months of sessions — gains a kind of self-knowledge that used to depend on memory and whatever made it into the notes. The practitioner gains it too.
Experience
The session itself: the human-to-human relationship, the moment a client's system softens in the presence of another person. Parts Companion is a silent observer here. It doesn't contribute observations during live sessions, and it isn't a synthetic therapist a client chats with to simulate therapy. None of this is ours to automate, and we have no interest in trying.
Integration
There's a saying among practitioners that the real change happens between sessions. A meditation generated from this week's session, for this client's parts, picking up exactly where the work left off, gets practiced — because it fits. Among clients whose practitioners use Parts Companion, 88% report continuing the work between sessions.
For most of their history, inner technologies have been one-size-fits-many: the same recorded meditation for every listener, the same workbook for every reader. These practices can now be tailored to one person, one week at a time — meeting people where they are and going as deep as they're ready to go.
"Just because we can doesn't mean we should."
That is not a mantra you'll hear much in Silicon Valley, but it's one we ask out loud and consider as a team, feature by feature. The capacity we guard most carefully is the client's own: their agency, and their right to their own self-discovery. These are the principles that run through every feature decision.
From parts, not about them.
What we write for a client comes from the client's parts, not from our analysis of them. A part is a member of the client's inner system with something to say, and the work is to receive it. So the content Parts Companion generates — a part's description on the map, a personalized meditation — is built from what the part actually said and showed in session, held in the client's own words, with no interpretation or analysis layered on top.
Even our reflection prompts keep this grammar: we ask what a client learned from a part, never about it.
The client directs. We supply the canvas.
When a client wants to externalize a part — an image of the protector they've been describing for weeks — the client directs its creation: describing, revising, regenerating until it matches what they see inside. We supply the canvas and the craft; the client supplies the vision.
Deliberate withholding.
There are places in the product where the system could pre-fill an answer it has inferred from sessions, and we hold it back on purpose: on a card the client is authoring about one of their parts, the system's guesses are never written in as if the client had said them, because a suggestion that arrives first quietly becomes the answer.
What the client authors and what the system inferred live on separate layers, and the client's layer always outranks ours. Everything we generate remains a draft the client can change. Only the client says what is true of their inner world.
The system suggests. People decide.
The same principle extends to the practitioner: consultation feedback is offered to grow a practitioner's own capacities, never to substitute for their judgment. We taught the system how a clinical consultant reads a session by collecting and studying real-world consultations from the most senior, accredited consultants in parts work — and then it suggests, while the practitioner decides.
Attunement for the clients served least.
Out of the box, models like these are least attuned to the clients who are already least served; the themes that Global Majority, LGBTQ+, and neurodivergent clients carry into parts work are exactly where general training runs thinnest. We decided early that this would not stand in our product. With Tamala Floyd, LCSW, we brought her framework of nine kinds of racism-related harm into Parts Companion and tested it against full-length practice sessions with practitioner missteps deliberately written in. With Sand Chang, PhD, we built the LGBTQ+ attunement the same way, and a round with neurodivergent clinicians is underway now.
Attunement is opt-in: a client tells Parts Companion what to expect in their sessions, never the other way around. No market forced any of this. We did it because technology this powerful should serve all people — read the full story.
Expensive in ways that "don't math" for a startup.
Doing this work properly costs more than a typical startup could justify, and that is precisely why a philanthropic actor belongs in this space. Several operating principles directly shape how Parts Companion is built.
HIPAA compliance from the first day
The same record-grade security as your EHR, verified by an independent security audit every year.
Session data never trains AI models
Not ours, not anyone's. Our AI vendors are contractually bound to this, too.
A measured footprint
We don't train models; we only use them, choosing the smallest model that can do each job well, and we track the cost of every session we process. Processing a session uses about as much energy as four minutes of streaming video. The numbers behind that live at partscompanion.org/environment.
A clinical team shapes everything
Lead trainers, approved consultants, and published clinicians shape and review what the system produces — meet the team.
An open door
We know practitioners carry real questions about AI — about privacy, about energy, about what belongs in the therapy room at all. We'd rather answer them than talk past them. Write to us at hello@partscompanion.org, or bring your community's questions and we'll show up for the conversation.
The work belongs to the two people in the room. We just help it travel well between sessions.