Psyche

Core personality, cognition, wellbeing, and motivation. Plus 10 written interview questions.

These ranges are worked out from the item counts, not measured from real sessions — we have not published an observed median yet, and we would rather show the arithmetic than a single confident number. Most of the spread is how long you spend writing. There is no timer, and you can leave and come back.

Not a clinical assessment or diagnosis. No account, and nothing you write is linked to an identity we hold — but your written answers are your own words, so please leave names and contact details out of them.

What is validated here, and what is not

The scales are established, published instruments with their own validity evidence. The report is not. The combined battery, the weighting, the interview synthesis, and the narrative you receive at the end are an experimental, interpretive layer built on top of those scales — it has not itself been clinically validated, and it inherits nothing from their validation.

Where your answers go

What we do not promise: zero retention covers the CONTENT of the request — your text, and the report written from it. It does not cover the billing and abuse-prevention records OpenRouter and the serving provider keep about the fact that a request happened; those are theirs, under their own policies, and we cannot promise anything on their behalf. The zero-retention endpoints for this model are published in both US and European regions and we do not pin which one a given request reaches.

If you use the Claude MCP route instead, your interview happens inside your own Claude conversation: Anthropic receives and handles that text under your own Claude subscription’s terms, not ours, and Claude — not GPT-5.6 Luna — writes the report, in that same conversation. What reaches Ashita Orbis through mcp.ashitaorbis.com is your ten written answers and nothing else: no scale scores, because that route does not administer the scales. Your answers are stored the same way as the ordinary flow once they arrive, and because no report is generated on our side they stay in the 7-day bucket — and you can delete them sooner from inside Claude, by asking it to use the delete tool with the session id and secret it gave you. Nothing on that route is sent to OpenRouter. One boundary we cannot cross for you: the answers you type to Claude exist in that conversation, under your own Claude account and its retention settings, before they ever reach us. Deleting the Psyche session removes our copy, not that one. Two smaller honesties about that route: what it records alongside your answers is a random session id, timestamps, and a record of what your Claude client asserted about consent — which is a record of the client’s assertion, not proof you were asked, because no MCP server can see its caller’s conversation.

Account-unlinked, and pseudonymous — not anonymous.

We never ask for a name, email, account, or device identifier, and your session is only a random ID. But ten open-ended answers are your own words: they can identify you if you put identifying details into them. That is pseudonymous processing, not structural anonymity, and we would rather say so than let the word "anonymous" do work it cannot do.

Please do not type names, contact details, employers, addresses, or anything that would identify another person. Write about the situation, not the roster.

How long any of it is kept

How much a score can give away, stated precisely rather than reassuringly: a scale score is a total over that scale’s items, so the fewer items a scale has, the more the score pins them down. The thinnest scales here have two items (the 60-item Big Five facets, and the locus-of-control pair), where the raw total fixes the sum of the two answers — and, only at the very top and bottom of the range, the two answers themselves. Nothing thinner exists in either tier: there is no one-item scale. The depression and anxiety screens are eight and seven items, so no individual answer to either is recoverable from what is sent.

The Psyche application stores no account, no device identifier, and sets no cookies. It does not store your IP address either — but the honest version of that sentence is longer. Your browser necessarily sends its address to Cloudflare to reach us at all. To stop one person exhausting the report budget, the server computes a short keyed HMAC token from that address and stores the token, never the address; the key is a server secret held outside the database, and the day is mixed into it, so tokens do not correlate across days and cannot be worked back to an address by anyone reading the database. The daily cleanup drops tokens once they are over 24 hours old, which means a token can exist for at most about 48 hours. Cloudflare, which hosts and shields the site, separately processes standard request metadata including IP addresses in order to serve the request, as it does for any site it serves; that metadata is not stored with, or linked to, your session. Deletion above removes rows from our database — Cloudflare’s own platform-level database backup and restore mechanisms sit underneath that and are governed by Cloudflare’s retention, not ours.