About

DharmAI takes a situation from your life or your practice and gives back counsel of the shape a well-read friend would give: a short reflection, the passages of the Pāli canon that bear on it, DharmAI’s own reflection on what you brought, and one thing to do. It remembers what you are working on.

It is not a teacher, not a therapist, and not a chatbot in the Buddha’s voice. It is a way of finding what the suttas actually say about what you are carrying, and being met where you are while you read it.

What it will not do

• It never speaks in the first person as the Buddha, and never speaks as, or names, a living or recently deceased teacher. The lenses are labelled emphases of a tradition, never persons.
• It is built to say nothing about what the Buddha taught or what the suttas say without a retrieved passage behind it. A validator drops any passage that was not retrieved for the reading. The models that write a reading see only DharmAI’s own summaries of the passages, never the translations, and every sentence of a reading is checked for claims about what the Buddha or the texts say: a passage note is removed when it claims more than DharmAI’s own description of the passage says, and any other sentence is judged against the descriptions of the passages on the page and removed when none of them says it. Two kinds stand without a passage: on a reading that finds none, the plain statement that the suttas do not speak to the question, and on a medical question, that they do not treat or cure conditions. That check reads our descriptions of the passages, not the passages, so it is weaker than reading the text itself. The section labelled “DharmAI’s reflection” is DharmAI’s own reading of what you brought and what you might try, in its own voice and never a teacher’s: it draws on the passages, and it is not what they say.
• It makes no clinical, diagnostic, or medical claims, anywhere, in its copy or in a reading.
• It does not continue past a crisis signal. Every turn is checked before and after generation, and a crisis replaces the whole response with a hand-off.
• It has no timers, streaks, badges, or points, and it will not send notifications.
• It does not use what you write to train anything, ever, by us or by a model provider.
• It will not generate new suttas, compose scripture, or pass a paraphrase off as the text.

What you write

What you write here is encrypted with a key that belongs to your account. We don’t train on it, our model providers don’t keep it, and you can export or delete all of it at any time. The only time your words leave our servers is to check and generate your readings, and they aren’t stored there. Nothing you write here is shared with anyone outside DharmAI except the AI services that check and write your readings, which are set not to keep it. If a safety check flags something you wrote, the person who runs DharmAI may open that one thread to review it, and every such view is logged.

In plainer detail: every situation, reading, follow-up and profile is sealed under a key that exists only for you, before it reaches the database. It is opened only inside the request that is serving you, and the plaintext is gone when that request ends. It is never in a log, a trace, an error report or an analytics row.

And the honest limit, because a promise you could catch us overstating is worth less than a smaller one that holds: we can technically decrypt, because the server has to read your words to generate the reading. What the design guarantees is that nothing is stored in the clear, that one key opens one person’s writing and no one else’s, that deleting is real, and that no provider keeps or trains on what passes through.

Export or delete everything whenever you like. Deleting removes your key, and nothing you wrote can be read again, by anyone.

Where the texts come from

Every passage you see is a verbatim segment of Bhikkhu Sujato’s translation, shown under SuttaCentral’s own segment ids, with the Mahāsaṅgīti Pāli beside it and a link that goes to the page on SuttaCentral. Those translations are dedicated to the public domain under CC0 by the people who made them, and the Pāli root text is in the public domain. We read them from the public bilara-data repository at a recorded commit, with the licence recorded per text. We do not crawl SuttaCentral, and we do not train or fine-tune any model on the texts.

SuttaCentral asks, beside that dedication, that the material not be used to build generative-AI datasets or the apps downstream of them. The request is a request and not a licence term, and we are not going to pretend otherwise — but someone asked, and we are doing a narrow version of it anyway, so we owe a clear account of what that version is. The model never produces the text: it selects passages, and the words shown as scripture come from the corpus. Nothing is paraphrased into scripture, nothing is fine-tuned, the passages are never put behind a paywall, and no endorsement is claimed or implied. Every passage carries its SuttaCentral segment id and links to Bhikkhu Sujato’s translation there, every page credits him by name, and the links send you to the source rather than keeping you here. Once there is revenue, a share of it will go to SuttaCentral as a recurring donation; nothing is charged during the beta, so there is none yet. If they publish a technical signal about how their data may be used, we will honour it; if they want the attribution worded differently, we will change it.

No SuttaCentral text, translation or Pāli, is sent to any model or embedded. The models that write a reading are given only DharmAI’s own description of each passage: a summary written for DharmAI, why it applies and where it stops. The passage you read is fetched from the corpus by its segment id, with no model involved. We control what we send, not what a model read in its training, and a reading that writes out a passage it was offered is rejected and written again. What the models write stays visibly apart from the text: the passages in their own setting, and DharmAI’s own reading under the label “DharmAI’s reflection”, never presented as what the suttas say.

Some people will still object to this existing. That is fair, and we would rather be criticised for doing it carefully than do it carelessly and hope nobody notices.

Reaching a person

Nothing here replaces a teacher, and some things should not be brought to a machine at all. When a reading meets one of those, it says so and stops.

To find a human teacher: most traditions keep public directories of monasteries, centres and teachers who take questions from practitioners, and many monasteries answer letters and run open evenings. If you are in crisis: in the US you can call or text 988 any time, and findahelpline.com lists free, confidential lines elsewhere. If practice itself has become destabilising, Cheetah House works specifically with meditators in difficulty. A doctor or a therapist is the right person for anything medical. This is not a crisis service and does not watch for one between visits.

Bring something, or see what is stored for you in settings.