A product we build and run
PenPaper AI
A personal agent that remembers, follows up and acts with permission: across web, desktop, mobile and WhatsApp, in India's languages.
A product StackHog built and operates itself, not a client engagement. Nothing on this page is a claim about work done for anyone else, and no usage or revenue figure appears because none has been published.
What it does
A little less to remember.
Reads what arrives
Mail, bills, statements and school messages become records and reminders without you typing them in.
Follows up on its own
Due dates, renewals, appointments and medicines are chased before they are late, not after.
Acts only with permission
Pays, books and fills forms when you allow it — once, or as a standing grant you can revoke.
Wherever you are
Web, Mac, Windows, Android, iPhone and WhatsApp, in English or India's languages — one memory across all of them.
How we built it
Permission, audit and speed are the architecture.
An agent that acts for you touches your inbox, your statements and your money. The hard part is not the model calls: it is permission boundaries, an audit trail for every action, and being fast enough on a phone to feel like a conversation.
Ingest, then structure
Mail, statements and chats become typed records — a purchase, a due date, a prescription — with the source page kept alongside every number.
Permission on every tool
Each action is a registered tool marked read-only, ask first, or standing grant, and a screening layer checks for tainted input before it runs.
A browser inside the agent
Where there is no API — a bill portal, a government form — the agent drives a headless browser under the same permission checks.
Fast, logged and capped
A classifier routes each turn to the cheapest capable model, prompts are cached, and every run is logged with its cost against quotas and a daily cap.
The same architecture, for you
We run this ourselves. We can build yours.
Next step
Bring us the agent you want to run.
If you are putting an agent inside a process that touches customers, money or regulated data, the permission model and the audit trail are the design — and they are the part we have already built once.