A council, not an assistant.

When you ask a question on Banyan, you are not asking an AI. You are convening a small council of historical figures — Marcus Aurelius, Rumi, Viktor Frankl, Pema Chödrön, Marianne Faithfull — chosen for the question you brought. Each one answers in their own register. They are allowed to disagree.

Every voice is grounded in a curated library of their own writing and recorded thought. The AI is not making them up from internet vapor; it is given their words and asked to speak in their direction. The result reads less like a chat window and more like a circle.

ChatGPT's memory is a feature it controls. Your Portrait is a feature you control. It is the slowly-growing record of what you have shared: journal fragments, what season you are in, what you have already wrestled with, who you have already grieved.

You choose what goes in. You choose, at the moment of each question, whether this conversation should draw from it. You can read it. You can edit it. You can delete a single piece, or burn the whole thing down. None of that is buried in a settings menu — it lives on your profile.

Banyan does call out to a language model to compose advisor replies — there is no way around that. What we changed is what we send.

When a thread draws on Your Portrait, every fragment passes through a redaction layer first. Names of people are replaced with [name]. Email addresses become [email]. Phone numbers, street addresses, URLs, long account-number-shaped digit strings — all stripped. The model receives the shape of your situation, not the index of your life.

The original text stays in your account, unredacted, exactly as you wrote it. The redaction only happens on the way out the door.

Every council sitting is saved as a thread under Me → Threads. You can come back tomorrow, next month, after a hard week, and pick up where the conversation left off. The advisors remember what was said. You remember where you were.

There is no "new chat" button you are pushed toward. The default is continuity. The deletion is on you.

Behind each advisor sits a hand-built library of their own teachings, letters, dharma talks, sermons, interviews. The AI retrieves the most relevant passages for your question and is instructed to stay close to them.

This is the difference between asking the open internet "what would the Buddha say" and being handed an actual stack of his teachings to read from. Fewer hallucinations. More of the real thing.

Banyan is not therapy. It is not a crisis line. And we do not pretend it is. Every single response from every single advisor passes through a safety layer that knows what to do if a question touches self-harm, abuse, or crisis: drop the literary register, name what was heard, and point to a real human resource — 988 in the US/Canada, 116 123 in the UK and Ireland, findahelpline.com elsewhere.

A poetic answer is not allowed to be in the way of someone who needs a person.

Most AI products are measured in time-on-app and tokens-per-user. Banyan is measured in whether you closed the laptop a little quieter than you opened it.

The interface is sparse on purpose. The voices are slow on purpose. The conversations end on purpose. The grove is not trying to be your everything — it is trying to be the bench under the tree.