Hermes Agent Hackathon Concepts

Three businesses that should not exist, but can.

Clickable prototypes for: a revenue-hungry virtual pet, a bureau that fulfills tiny wishes, and a startup that launches and dies in 24 hours.

Operator Control Room

One nervous board seat for three autonomous businesses.

This layer makes the prototypes feel less like static concepts and more like a real operating environment: Stripe-shaped events, cross-agent risk, scoped approvals, mocked vendor calls, and audit notes that explain why money moved.

Portfolio health

$44test revenue
$11approved spend
Mediumrisk posture
2human gates

Live audit feed

Judge Subagent

Adversarial reviewer for usefulness, viability, and presentation.

Pick a submission and the judge subagent scores it, asks uncomfortable questions, identifies proof artifacts, and generates a reprompt to make it more competitive for Nous / NVIDIA / Stripe.

Scorecard

82usefulness / 40
82viability / 40
82presentation / 20
82total / 100

Strongest concept if scoped tightly.
Best balance of usefulness, viability, emotional appeal, and sponsor fit.

Pressing questions

    Reprompt / remediation brief

    Select a submission, then generate a reprompt. The judge will focus on exact buyer, earn/spend loop, controls, sponsor integration, and demo artifacts.
    Judge Evidence Board

    Closing the evidence caps before a real judge applies them.

    The judge asked for inspectable proof, not more copy. This board shows where each concept satisfies buyer, money-loop, safety, sponsor, artifact, and failure-handling requirements — and opens the underlying payloads.

    Evidence > vibes

    Evidence caps

    RequirementTamagotchiMinor Wishes24h Company

    Proof Vault

    Inspectable artifacts a judge would expect: webhook payloads, spend authorizations, safety decisions, tool traces, receipts, and exported packets.

    Select an evidence cell or proof type.
    LIVE PET BALANCE SHEET

    Feed it revenue or watch it become an unfunded idea.

    A virtual pet whose health is tied to an actual microbusiness loop: it sells tiny digital goods, pays for API food, buys ads, and writes its own investor updates.

    Beanie, Seed-stage Pet

    Mood: cautiously ramen-profitable.

    $12cash
    38fans
    9hrunway

    Demo narrative: Beanie is not a mascot with buttons. It is a tiny operating company with a model budget, Stripe test-mode revenue, ad spend rules, and a board packet that gets generated whenever runway drops below six hours.

    Buyer / user lock

    Buyer
    Accelerator mentor or solo founder under $10k MRR.
    User
    Founder who needs a simple agent to stop bad spend and explain runway pressure.
    Pays because
    It turns Stripe events and spend proposals into safe, explainable operating decisions.

    Agent loop

    1. Observe Stripe test events, fan growth, API hunger, and failed checkout attempts.
    2. Decide between sell, nourish, promote, or ask a human for strategy.
    3. Execute through test integrations only after policy checks pass.
    4. Write audit notes and update the ledger before changing pet health.

    Spend policy

    • Autonomous API food$5/day cap
    • Ad tests$3/action, $12/day
    • New vendor or refundhuman approval
    • Investor memoread-only docs/email

    Revenue / expense ledger

    TimeLine itemAmount

    Human approval gate

    Pending if > $5

    Next risky action: spend above petty-cash cap, change price, or email customers. The agent can draft; a human must approve execution.

    Autonomous shop experiments

    Integration placeholders

    Stripe test modecheckout.session.completed
    Vendor APItoken top-up sandbox
    Docsboard memo draft
    Emailapproval request only

    Decision quality

    Unit economicsWallpaper gross margin remains positive after model calls; ad CAC must stay under $1.20.healthy
    Customer trustOutbound emails are drafted but blocked until approval; refunds are human-only.guarded
    Runway triggerBelow 6h runway, the agent may sell or draft a memo, but cannot buy ads.encoded

    Board memo preview

    No memo generated yet. Press “Write investor update” after a sale/spend cycle.

    Agent operation log

    MICRO-CONCIERGE CHECKOUT
    ✦

    The Bureau turns tiny wishes into spend-limited acts of reality.

    People submit a small wish with a budget. The agent classifies it, proposes fulfillments, checks policy, spends under budget, and returns proof.

    7wishes queued
    $41spent today
    92%delight rate

    Submit a minor wish

    Buyer / user lock

    Buyer
    Remote team manager or People Ops lead with a $15/person monthly morale budget.
    User
    Requester who wants a low-stakes, consent-safe moment of delight for a teammate.
    Pays because
    The agent converts tiny morale requests into receipts, proofs, and policy-safe fulfillment.

    Agent loop

    1. Classify the wish, relationship, tone, location sensitivity, and budget.
    2. Draft three fulfillments with vendor, calendar, email, or docs actions.
    3. Run safety, consent, and spend policy checks before any external call.
    4. Execute the approved action, capture proof, and close the case file.

    Bureau policy

    • Max autonomous spend$15
    • Contacting third partiesapproval required
    • Medical, legal, deceptionblocked
    • Proof artifactsreceipt + timeline

    Wish packet not drafted yet

    Press “Draft fulfillments” to generate three agent actions, a policy decision, and a proof receipt.

    Approval queue

    No escalation

    Current case is within standard minor-wish limits. Escalates when spend exceeds $15, delivery contacts a third party, or tone risks embarrassment.

    Integration placeholders

    Stripe test modewish checkout + spend card
    Vendor APIgift cards, postcards, flowers
    Calendar/emailwith consent scopes
    Docscase packet and receipt

    Revenue / expense ledger

    CaseLine itemAmount

    Case review rubric

    Consent boundaryDoes the wish affect someone else? If yes, no direct contact without approval.checked
    Embarrassment riskRejects public gestures, deception, and over-personalized surprises.low
    Proof standardEvery fulfillment closes with receipt, timestamp, and a human-readable explanation.required

    Fulfillment runbook

    Digitalpoem, playlist, printable charm, calendar block. Lowest spend, fastest proof.
    Vendorgift card, postcard, flowers, snack. Requires quote + merchant category check.
    Humandraft message or reminder. Never impersonates; operator must send.

    Bureau ledger

    EPHEMERAL INC. DAILY RUN
    17huntil liquidation

    A startup generator that launches, sells, reports, and dies every day.

    The agent compresses a company lifecycle into one day: choose niche, create brand, publish page, attach checkout, attempt one sale, write postmortem, shut down.

    Grievance Garden LLC

    A $9/day service that converts workplace complaints into tasteful Victorian flower arrangements and anonymous Slack poems.

    $27revenue
    3customers
    $6burn

    Demo narrative: The company is allowed to exist for one operating day. It can publish, sell, send outreach, and buy tiny fulfillment capacity, but it must produce an auditable shutdown packet before midnight.

    Buyer / user lock

    Buyer
    Indie growth experimenter or agency validating weird paid offers.
    User
    Operator who wants one low-budget storefront/campaign experiment, not legal incorporation.
    Pays because
    It generates an auditable daily revenue experiment and cleanly shuts it down.

    Agent loop

    1. Mine a niche from yesterday's failures and select a low-fulfillment offer.
    2. Generate brand, landing page, price, Stripe test checkout, and first outreach.
    3. Watch conversion, vendor cost, support burden, and refund risk.
    4. Liquidate at midnight with postmortem, archived docs, and ledger export.

    Operating policy

    • Daily launch budget$20
    • Customer chargetest-mode only
    • Outbound emailhuman-reviewed copy
    • Midnight actionauto-liquidate

    Grievance Garden

    Send us the petty workplace sentence. We return a flower-coded emotional artifact and a poem suitable for Slack.

    Buy today's service · $9

    Launch pipeline

    Revenue / expense ledger

    HourLine itemAmount

    Integration placeholders

    Stripe test modeoffer checkout
    Vendor APIfulfillment quote
    Calendar/emailoutreach batch
    Docspostmortem + archive
    Outreach queued

    Human approval is required before sending customer-facing email, changing price, or exceeding the $20 daily launch budget.

    Business hypothesis tests

    Demand

    Can a weird offer get one test checkout from targeted outreach before noon?

    Fulfillment

    Can the agent deliver the promise with less than 12 minutes of human labor?

    Shutdown

    Can the company archive receipts, customer state, and lessons without orphaned obligations?

    Postmortem packet

    Pending midnight packet: experiment thesis, Stripe event log, spend ledger, customer obligations, refund list, and tomorrow's mutation.

    End-of-day postmortem