Why not just ChatGPT?
A fair question with an honest answer. Sathi is not trying to be a smarter chatbot — it is trying to be a different layer of computing: the persistent, private intelligence layer of a person's and family's life.
This layer is kept in English so it stays current; the full story lives at /principles in Nepali, English, Japanese, and Hindi. Machine map: /llms.txt. Everything here is true as of 2026-08-20; plans are labeled as plans.
What generic assistants do better — today
Honesty first, because every other claim on this site depends on it: frontier hosted models are stronger raw reasoners than the open-weights models Sathi runs on today — we accept that gap in exchange for independence, and say so on /principles. They also have broader third-party integrations, more polish, and the economics of a giant. If you need a maximally capable general oracle with no memory of your life required, they are excellent — and Sathi can even rent frontier capability for specific tasks later, behind the same disclosure rules.
Where Sathi is a different thing entirely
| A generic assistant | Sathi | |
|---|---|---|
| Persistent personal context | Session memory features, provider-owned | The core architecture: a bi-temporal claims store built from your own archive |
| Where the memory lives | On the provider's infrastructure, under their policy | On one deployment's own server; the model provider sees per-request context, never the store |
| Provenance | Answers rarely say where a memory came from | Every claim carries its source, when it became true, and how confident it is |
| Temporal memory | Latest-write-wins, mostly | Supersede, never overwrite — "what did I believe last year?" stays answerable |
| Family context | Separate accounts, no shared boundary | Permission-structured from day one: shared where chosen, private by default, consent on the graph |
| Your archive | Export varies; deletion policies vary | Total signed export, provable deletion, a signed receipt when you leave — tested code paths |
| Devices | Separate products with separate accounts and clouds | One Sathi, many organs: a device adds senses, never a second brain (speaker experimental today) |
| Model independence | You are the model's customer | Models are replaceable parts behind abstractions; your understanding survives every swap |
Statuses for every capability, honestly tensed, live on /capabilities.
The one-sentence version
A mediocre assistant with extraordinary continuity beats an extraordinary assistant with mediocre continuity — because the continuity is yours, and it compounds. Sathi's bet is that reasoning becomes a commodity and the substrate — your archive, your understanding, your permissions — is what remains scarce and remains yours.
See it rather than believe it
The principles page shows five real moments and one full belief revision — old belief, new evidence, what Sathi does differently. Or just try it: guest chat is open, needs no account, and remembers nothing.