🤫husshhussh
  • Wiki
Reserve
🤫husshhusshOneOne Puppy
🤫 Build · the Blueprints

Blueprints for purposeful agents. Then bring your own ingenuity.

This is the deep, buildable reference a vibe coder starts from and keeps coming back to. Concrete, purposeful agents you can build for 🤫 Agent One, each specified end to end so nothing is left to doubt, yet loose enough that your own human ingenuity is the point. Build the smallest slice, keep a few honest checks, ship it to one real person, then show it to the world and bring people along.

The best practicesThe field guide
The one rule under all of it

It is all about show and tell in the world.

The board says this with serious candor, so remember it. We do not win by describing a future. We win by showing a real, working thing to a real person, honestly, and bringing them along. A blueprint is not finished when the code runs. It is finished when someone who was not in the room can see it work, understand it, and want it.

Show

Build the smallest thing that actually works, then let a real person watch it do one real job, end to end, with the receipt. Working beats promised, every time.

Tell

Explain it in plain words anyone can follow. If you cannot say what it does and who it helps in one honest breath, it is not ready to show.

Bring people along

Build in the open. Publish what changed, credit what you learned from, and let others copy the good parts. A good idea spreads when everyone can build to it.

How to read a blueprint

10 blueprints, across 6 parts of a life.

Each blueprint is a starting line, not a cage. It gives you the purposeful job, who it helps, the smallest slice to ship first, the consented context it needs, the rails it reaches for, the guardrails that keep the human in control, a few honest evals, and, most important, where your own human ingenuity matters most. Pick one that solves a real need in your life or your market, and go deep.

Blueprint · Personal

🤫 Morning Brief

Assemble one honest picture of the day ahead, so the person starts clear instead of scrambling.

Who it helps: Anyone whose morning is a scramble across five apps: the busy professional, the parent, the owner.

Ship this first · the smallest slice

Read today's calendar and the small handful of messages that truly need a reply, and produce one short brief: what matters today, who is waiting on you, and the one thing to do first.

Consented context it needs
  • Today's calendar (read-only, today's window only).
  • The inbox, scoped to unread and flagged, not the whole archive.
  • The person's stated priorities for the week, if they set them.
Rails it reaches for
  • MCP to read calendar and mail tools.
  • No writes in the first slice. Reading only, so trust is earned before it ever acts.
Guardrails · keep the human in control
  • Least privilege: read today's window, not the person's history.
  • A receipt for every source it read, visible in the brief itself.
  • Never sends anything. It proposes; the human decides.
Evals · how you know it is good
  • Does the brief name the genuinely important thing, not just the loudest?
  • Would the person have missed something without it? Ask them for a week.
  • Is it short enough to read before coffee? If it needs scrolling, it failed.
Where your human ingenuity matters most

Deciding what 'important' means for this person. That judgment is yours, and it is the whole product.

Blueprint · Personal

🤫 Inbox Triage

Sort the flood down to the few messages that are genuinely the person's to decide, and draft the routine replies.

Who it helps: Anyone facing a few hundred messages a day where most are noise and a few truly matter.

Ship this first · the smallest slice

Label incoming mail by what needs the human, draft replies to the clearly routine ones, and surface the short list that is genuinely theirs. Send nothing without approval.

Consented context it needs
  • The inbox, scoped to incoming and unresolved.
  • The person's own past replies, to learn their voice, held by their Agent One.
Rails it reaches for
  • MCP to the mail tool for read, label, and draft.
  • A2A only if a reply needs another agent's input, and only by consent.
Guardrails · keep the human in control
  • Drafts, never sends, until the person turns on auto-send for a specific, narrow category.
  • Every label, draft, and archive is a receipt the person can undo.
  • Never reads or acts on an account the person did not connect.
Evals · how you know it is good
  • Precision on the short list: is every surfaced message actually worth the human's time?
  • Do the drafts sound like the person, or like a robot? Read three aloud.
  • How many important messages slipped through? That number must trend to zero.
Where your human ingenuity matters most

The taste to tell a real message from a well-disguised one, and to write a reply that sounds like a person who cares.

Blueprint · Household

🤫 Family Logistics

Keep a household's moving parts, schedules, pickups, and reminders, from colliding.

Who it helps: The parent or household manager holding a family's logistics in their head.

Ship this first · the smallest slice

Watch the shared family calendar, flag the collisions and the gaps (who covers the pickup when both parents have a meeting), and propose the fix before it becomes a crisis.

Consented context it needs
  • The shared family calendar, by consent of each adult.
  • The known standing commitments: school hours, practices, work blocks.
Rails it reaches for
  • MCP to the calendar tools of each consenting member.
  • A2A between family members' Agent Ones, so no one central app holds the whole family's life.
Guardrails · keep the human in control
  • Each person shares only their own calendar, and only the busy or free signal if they prefer.
  • Proposes reschedules; a human confirms every one.
  • A receipt to each member for anything read or suggested about their time.
Evals · how you know it is good
  • Did it catch the collision a tired parent would have missed?
  • Are its proposed fixes actually workable, or naive? Watch a real week.
  • Does it reduce the mental load, measured honestly by the family, not by a dashboard?
Where your human ingenuity matters most

Understanding a family is not a scheduling problem, it is a trust and fairness problem. Design for that.

Blueprint · Small business

🤫 Invoicing

Raise invoices, send them on time, and chase the unpaid ones politely, so the owner stops doing it at midnight.

Who it helps: The freelancer, the shop, the small-business owner billing by hand.

Ship this first · the smallest slice

From the owner's agreed terms, draft each invoice, queue it for approval, send on schedule, and send one polite reminder on the day it is due.

Consented context it needs
  • The client list and agreed terms, held by the owner's Agent One.
  • The record of what has been delivered and is billable.
Rails it reaches for
  • MCP to the invoicing or accounting tool the owner already uses.
  • AP2 and UCP are how settlement will flow, when live. Today the agent surfaces and organizes; it does not move money.
Guardrails · keep the human in control
  • Bills only the clients and terms the owner authorizes.
  • Every invoice and reminder is approved or auto-approved within a narrow rule the owner sets.
  • A full trail: what was billed, to whom, when, and why.
Evals · how you know it is good
  • Did it bill the right amount to the right client, every time? Zero tolerance for a wrong number.
  • Are the reminders polite enough to keep the relationship? Read them.
  • Days-to-paid: does it shorten honestly over a quarter?
Where your human ingenuity matters most

Knowing which client gets a gentle nudge and which gets a firmer one. Money and relationships are human. Keep them so.

Blueprint · Money

🤫 Receipt Wrangler

Turn the shoebox of receipts into a clean, categorized, month-end report the person just reviews and submits.

Who it helps: Anyone who dreads expense reports: the traveler, the contractor, the small team.

Ship this first · the smallest slice

Capture each receipt as it happens, categorize it, match it to the card charge, and assemble the finished report ready to submit.

Consented context it needs
  • The receipts the person connects, photo or forwarded.
  • The card feed, read-only, scoped to the reporting window.
Rails it reaches for
  • MCP to the receipt store and the card feed.
  • Export to the expense tool over its API, on approval.
Guardrails · keep the human in control
  • Reads only the receipts and card feed the person connects.
  • Submits nothing without sign-off, or within a narrow auto-rule the person sets.
  • A receipt for every categorization and match, easy to correct.
Evals · how you know it is good
  • Categorization accuracy: does the hundredth receipt match the first?
  • How many receipts go missing? The answer should be none.
  • Time saved at month-end, measured against the old way.
Where your human ingenuity matters most

The edge cases: the split bill, the personal-and-business meal, the foreign charge. Your rules for the messy 5% are the craft.

Blueprint · Care

🤫 Care Check-in

Give a caregiver quiet reassurance that a loved one is okay, without hovering or surveilling.

Who it helps: The adult child of an aging parent, the parent of a young adult, anyone caring at a distance.

Ship this first · the smallest slice

On a gentle cadence the loved one consents to, confirm presence and a simple 'all is well,' and surface only what actually needs attention.

Consented context it needs
  • Presence and, if consented, location on the loved one's terms.
  • Pairs with 🤫 Tag One where the family uses it. Tag One is a safety and wellness wearable, not a medical device.
Rails it reaches for
  • MCP to the presence signal.
  • A2A between the caregiver's and the loved one's Agent Ones, so no company sits in the middle of a family's care.
Guardrails · keep the human in control
  • The loved one controls what is shared, to whom, and can revoke instantly.
  • Presence by default, precise location only if they choose it.
  • Every check-in leaves a receipt both sides can see. Care, not surveillance.
Evals · how you know it is good
  • Does it reassure without intruding? Ask both the caregiver and the loved one.
  • Does it surface a real concern promptly, and stay quiet otherwise?
  • Does the loved one feel respected, or watched? That answer is the whole test.
Where your human ingenuity matters most

The line between caring and controlling is human and delicate. Designing for dignity is the hardest and most important part.

Blueprint · Money

🤫 Renewal Watchdog

Catch every subscription and policy renewal before it silently charges or lapses, and put the choice in front of the person in time.

Who it helps: Anyone bleeding money to auto-renewals they forgot, or caught out by a policy that lapsed.

Ship this first · the smallest slice

Track every renewal date, surface the decision ahead of time with what changed, and renew, cancel, or flag for renegotiation on the person's instruction.

Consented context it needs
  • The subscriptions and policies the person adds, with their renewal dates and terms.
Rails it reaches for
  • MCP to the calendar and any billing tools connected.
  • UCP and AP2 will carry the action when settlement is live. Today it surfaces the choice; the person acts.
Guardrails · keep the human in control
  • Tracks only what the person adds, acts only on their decision.
  • Nothing renews or cancels without the person's word, or a narrow pre-set rule.
  • A receipt for every date watched and every action proposed.
Evals · how you know it is good
  • Did it catch every renewal, with enough lead time to actually decide?
  • Dollars saved from cancelled forgotten subscriptions, counted honestly.
  • Did it ever act without permission? That must never happen.
Where your human ingenuity matters most

Framing the choice so the person decides well in ten seconds. Good defaults and a clear 'what changed' are your craft.

Blueprint · Small business

🤫 Meeting Prep

Walk the person into every meeting already prepared: who, why, what happened last time, and what to decide.

Who it helps: Anyone with back-to-back meetings and no time to prep between them.

Ship this first · the smallest slice

Before each meeting, assemble a one-screen prep: the attendees, the last exchange, the open items, and the decision on the table.

Consented context it needs
  • The calendar event and its attendees.
  • Past threads and notes with those people, held by the person's Agent One.
Rails it reaches for
  • MCP to calendar, mail, and notes.
  • A2A to a counterpart's Agent One only where both consent to share an agenda.
Guardrails · keep the human in control
  • Reads only the person's own history with the attendees.
  • Shares nothing outward without explicit consent.
  • A receipt for each source pulled into the prep.
Evals · how you know it is good
  • Did the prep make the meeting sharper? Ask after a week of real meetings.
  • Is it one screen, or a wall? Brevity is the feature.
  • Did it surface the one thing the person had forgotten?
Where your human ingenuity matters most

Knowing what actually matters walking into a room. The relationship context, the unspoken decision. That reading is yours.

Blueprint · Creator

🤫 Content Repurposer

Turn one piece a creator makes into the honest set of formats their audiences want, in the creator's own voice.

Who it helps: The creator, the founder, the small team publishing across places and short on time.

Ship this first · the smallest slice

From one long piece the creator approves, draft the shorter cuts for each channel, in their voice, and queue them for the creator's edit before anything posts.

Consented context it needs
  • The creator's own work and past posts, to learn their voice, held by their Agent One.
  • The channels they publish to and each one's shape.
Rails it reaches for
  • MCP to the drafting surface and, on approval, the publishing tools.
Guardrails · keep the human in control
  • Drafts only. The creator edits and approves every post.
  • Uses only the creator's own consented material, never scraped or borrowed work.
  • A receipt for each draft and each post, easy to pull back.
Evals · how you know it is good
  • Does it sound like the creator, or like a content mill? The creator is the judge.
  • Does it respect the source, or distort it for clicks? Honesty over reach.
  • Does it save real hours without cheapening the work?
Where your human ingenuity matters most

Voice is human. The agent drafts; the creator's taste, humor, and judgment are what people actually followed them for.

Blueprint · Personal

🤫 Knowledge Librarian

Turn the person's own scattered notes, docs, and saves into a private brain they can actually ask.

Who it helps: The researcher, the professional, the lifelong learner drowning in their own saved material.

Ship this first · the smallest slice

Index the documents and notes the person connects, and answer their questions from that private material, with a citation to the source every time.

Consented context it needs
  • The documents, notes, and saves the person connects, held by their Agent One.
  • Pairs with the Personal Knowledge & World Model. See /research/personal-world-model.
Rails it reaches for
  • MCP to the person's document and note stores.
Guardrails · keep the human in control
  • Reads only what the person connects; nothing leaves their control.
  • Every answer cites its source in the person's own material.
  • A receipt for what was read to answer each question.
Evals · how you know it is good
  • Are the answers grounded in the person's real material, with a source, never invented?
  • Does it say 'I don't have that' honestly, instead of guessing?
  • Does the person trust it enough to rely on it? Trust is the metric.
Where your human ingenuity matters most

Deciding what a good, honest answer looks like for this person's work, and refusing to let the agent bluff. Grounding is craft.

The spine

The ten best practices every agent on 🤫 One is built to.

Short, memorable, and non-negotiable. Whatever you build, build it to these. The last one is the one that keeps you human.

01

Start with the human, not the feature

Name the real job a real person is trying to get done before you name a tool. Work backwards from the human. If you cannot say who this helps and how, do not build it yet.

02

Ship the smallest end-to-end slice

Not the platform, the thinnest path that helps one person, one time, all the way through. A slice you can ship this week and honestly evaluate beats a grand plan you cannot.

03

Consent-first, by construction

Touch personal data only through PCHP: scoped, purpose-bound, expiring, with a receipt. Ask for one field, one moment, one purpose. Never build a shadow copy of a person's life.

04

Least privilege, always

Request the narrowest scope that does the job, and the shortest expiry. Read today's calendar, not the person's history. Every extra permission is a liability you now owe them.

05

Leave a receipt for everything

Every access, every action, written to a ledger the owner can read and revoke. If your agent does something the person cannot see and undo, you built it wrong.

06

Draft, do not send, until trust is earned

Default to proposing, not acting. Let the person turn on auto-action only for a narrow, specific category once they trust it. Autonomy is granted, never assumed.

07

Be honest about live versus roadmap

Where money moves, say plainly: the agent organizes and surfaces today; settlement over AP2 and UCP, and the 🤫 Gold ID, are roadmap. Never let an agent imply a power it does not have.

08

Keep a few honest evals

Two or three real checks that prove the slice does what you promised beat a wall of vanity metrics. When the model changes, your evals tell you the truth. Never ship a personal-data flow you cannot verify.

09

Code like bacteria

Small, modular, self-contained pieces anyone could copy out without importing your whole world. Build a bigger backbone only where complexity truly demands it. Prefer 5 lines over 50.

010

Keep your human ingenuity

The blueprint gives you the skeleton. The judgment, the taste, the care for the specific human, that is yours, and it is the whole product. Let the machine do the typing; you bring the wisdom.

How we build it at 🤫

Story aligned to code, shown in the open.

We try to practice what these blueprints preach: small, self-contained code anyone could copy out; consent-first by construction; a receipt on every access; and a public story that maps to real, legible code. We publish our protocol openly, keep our research in the open, and show our work as we go, because show and tell is how you bring the world along.

Pick one. Build the smallest slice. Show it to the world.

Choose a blueprint that solves a real need, ship the thinnest end-to-end slice this week, keep a few honest evals, and let a real person watch it work. Then tell the story, and bring people along.

The field guideThe deep playbook: Building on 🤫 OneThe 🤫 Academy (train to build)Developers & open rails

One is a product of Hushh Technologies Corporation (brand: 🤫 “hussh”), an independent company. One runs on third-party silicon, systems, and cloud; all company names are used solely to describe the platforms on which One software runs. Hushh Technologies is not affiliated with, endorsed by, sponsored by, or partnered with any company named.