When one MahatOS learns how to do a task, every other MahatOS can learn the method — without any of them ever seeing your personal information.
Collective Skill Intelligence
This is MahatOS's compounding advantage — inspired by federated learning (a technique where devices share what they learned, never the raw data they learned it from), applied to everyday task automation.
One MahatOS learns a task
A user's MahatOS works through a multi-step task — say, filling a scholarship application form — the slow way, step by step.
The pattern is anonymized
MahatOS strips out everything personal. What remains is only the structure: “fill the name field, then the date-of-birth field, then click submit.” Never the name. Never the date. If a field isn't recognized, the system refuses to save it at all — it fails closed.
The user approves sharing
Nothing is shared by default. The user sees exactly what would be uploaded — the bare pattern — and explicitly opts in. Twice: once to save, once to share.
Every other MahatOS gets faster
The next MahatOS facing the same kind of form skips the trial-and-error and completes it in fewer steps — measured, not estimated.
The measured evidence
In our end-to-end test, a fresh MahatOS completed a mock scholarship form in 9 steps. A MahatOS that reused the shared pattern completed the same form in 6 steps — a one-third reduction, verified by the federation test in our regression suite. Small today, compounding forever: every skill any MahatOS learns makes the whole network faster.
steps, first-ever attempt
steps, reusing a learned pattern
Don't take our word for it — read the skill file
This is the actual saved skill from that test, exactly as stored. Look at what's in it: field roles like full_name and date_of_birth — and look at what isn't: no name, no date, no phone number, no ID. The structure of the task, none of the substance. The anonymization_verified: true flag means the system checked this before saving.
Real capture: the stored skill for the mock scholarship form. Roles only, values never.
The opt-in card shown before any skill is saved. Default is off.
Technical provenance
The privacy-preserving approach behind this system builds on a published Indian patent (Application No. 202541122148) on federated learning held by Tranquil AI, the founder's earlier venture. We cite it here as founder credibility and technical provenance — the patent belongs to Tranquil AI, not to MahatOS. The point isn't ownership; it's that privacy-preserving learning was built into this team's DNA from day one, not bolted on.