[ AI gateway · Robinhood Chain · $EMBRR ]
Every prompt burns supply.
One crypto balance. Every AI model. The fee buys the token and destroys it.
No subscription. No card. Withdraw what you do not spend.
- X · soon
- GitHub · soon
- Docs
- App · soon
[ Fund · Spend ]
One balance. The meter runs down as you work.
01 · Fund
Crypto in
Deposit from the wallet you already use, any supported asset, any supported chain. 1 USD becomes 1,000 credits. No card, no bank, no waiting on a transfer to clear.
02 · Spend
Any model, metered
Text, code, image, video and audio on one balance, metered at the published upstream rate plus a 10% platform fee. No subscription and nothing recurring.
[ Burn ]
The fee goes into the furnace.
Half of every fee is queued, and every 50 USD or one hour it buys $EMBRR on the open market and destroys it in the same transaction. Supply only moves one way.
- Revenue only, never treasury
- Bought on the open market
- One transaction per 50 USD or hour
- Every burn published
[ What is left ]
Supply only moves one way.
Every paid request retires a little more of a fixed supply. No distributions, nothing to stake, nothing to claim — the token simply gets smaller each time somebody uses the product.
Burned to date: 0 $EMBRR. The gateway is pre-launch, so there is nothing to burn yet.
Embrr AI
Session receipt
The gateway is pre-launch, so this is an illustration of the format. Real receipts begin in Phase 1 and carry a transaction hash from Phase 2.
Burned to date
0 $EMBRR
nothing to burn yet
Token
Not launched
launches in Phase 2, not before
Gateway
Pre-launch
Phase 0 — validation
Team allocation
0%
no presale, no allocation
$EMBRR · contract address · Robinhood Chain
Pre-launchTo be announced.
The token launches at the end of Phase 2, after the burn engine is live and the public burns page already carries real transactions. Until an address is printed in this box, any contract claiming to be $EMBRR is a fake.
[ 01 ] The split
Where your money actually goes.
We only ever split our own margin — never your payment. The model cost passes straight through to the upstream provider; the 10% on top is the entire company, and this is what happens to it. Move the numbers.
Did the request use a marketplace prompt?
Share of supply you hold
Holding discount 0% + usage discount 10% (earned automatically from monthly spend) = 10% off the fee.
What you pay
The fee, zoomed in
Your discount takes the fee from 10% to 9% — and it shrinks the burn with it. That is exactly why the cap is 40% and not 100%.
A prepaid AI gateway on Robinhood Chain. Fund one balance with crypto, spend it across text, code, image, video and audio models at published rates, and withdraw whatever you do not spend. A 10% platform fee sits on top of the model cost. Half of that fee buys $EMBRR on the open market and burns it; 30% pays the author of the prompt you ran; 20% runs the company.
[ 02 ] Positioning
The same product on the surface. The opposite underneath.
Incumbent AI gateways
Embrr AI
Card or bank required
Fund from the wallet you already use
Monthly subscription
Pay per request, nothing recurring
Balance locked forever
Unspent balance is withdrawable
Token = discount on a fee that does not exist yet
Token = destroyed by every fee that is actually charged
Catalog built one vendor at a time
Full catalog on day one through a licensed upstream aggregator
Team allocation, unlocked
No allocation, no presale
"Locked" with no locker
Locked in a contract, or not claimed at all
[ 03 ] The problem
Four reasons the category keeps disappointing.
01
Crypto users cannot pay for AI
The major AI platforms take cards and bank transfers. A large share of crypto-native users have no card that works across borders, or will not attach one. They hold liquid assets and still cannot buy inference.
02
AI tokens have no connection to AI usage
The usual pattern is a token that grants a discount on a platform fee. When the platform charges no fee, the token discounts zero. Usage climbs, the treasury fills, holders receive nothing.
03
Prepaid balances are one-way doors
Credits that can never be withdrawn are not a balance, they are a donation with a delay. The float sits on the operator's books and the user carries all of the counterparty risk.
04
Catalogs are advertised, not delivered
"Every AI model" is a headline. The catalog underneath is often a handful of endpoints, several listed but not runnable, with whole modalities empty.
[ 04 ] The burn
Four rules that never bend.
A burn is only worth something if it cannot be staged. These are the constraints that make ours checkable.
1
Revenue only, never treasury
No revenue means no burn. A burn paid for out of our own reserves is theatre: it moves our money in a circle and calls the circle demand. We will not do it, including on a slow week when the chart would like us to.
2
Buy first, then destroy
We hold no token inventory to burn. Each burn is a real market buy followed by burn(). Destroying tokens already in our own wallet would shrink our bag and place no bid.
3
Batched, not per request
A transaction per request is waste. The queue accumulates to 50 USD or one hour, whichever lands first, then fires once.
4
Published or it did not happen
Every burn writes a row on the public burns page: amount, tokens destroyed, transaction hash, block. No aggregate-only dashboards, no rounded monthly totals.
[ 05 ] The free tier
The free tier burns nothing. On purpose.
Three requests, no wallet, no signup, inexpensive models only. A free request costs us money upstream and earns none, so burning on top of it would be paying twice to fake a metric. The free screen still shows the machine working, honestly.
- The first line is a labelled simulation, never presented as a burn.
- The second is other people's real burns, each with a link to its transaction.
- Watching strangers burn is a stronger pitch than a fake counter, and it is true.
free tier · request 2 of 3
[ Sample screen ]Simulation — labelled as one
On a funded balance this request would have burned ~1.4 $EMBRR
Live feed — honest zero until the gateway opens
Burned in the last 5 minutes: 0 $EMBRR across 0 requests
[ 06a ] Upstream
Upstream through an aggregator, not vendor by vendor
We route through a licensed model aggregator instead of contracting each model vendor directly.
- Resale rights
- Several frontier vendors restrict reselling raw API access. An aggregator that already holds resale terms keeps us inside them.
- Catalog on day one
- Hundreds of models at launch instead of a handful added one integration at a time. This is the most visible gap in the incumbent product.
- No vendor lock
- Model deprecations and price moves are absorbed upstream instead of breaking us.
[ 06b ] Withdrawal
Withdrawal, and the rule that removes the risk
Unspent balance is withdrawable. Two constraints make that safe rather than clever.
- Same wallet only
- Funds return to the exact address that deposited them. Never a second address, no exceptions, no support override. There is no path a mixer could use, because in and out are the same key.
- Unspent only
- Consumed credits are gone. You are withdrawing a balance, not reversing a purchase.
[ 07 ] Who uses it
Four people, four journeys.
Primary — pays the bills
A developer or operator who already spends on AI every month.
step 01
Lands from a burn receipt somebody posted
step 02
Types into the box on the homepage. No signup. The answer streams back in seconds
step 03
Sees the labelled simulation line and the live global burn feed
step 04
The fourth request hits the wall. Connects a wallet
step 05
Funds with whatever is already in the wallet. Reads the word withdrawable
step 06
Works. Switches models without a second subscription
step 07
Ends the session with a burn receipt card, one click to share
[ 08 ] The proof pages
Published, or it did not happen.
Two public pages, no login and no API key: every burn as its own transaction, and a solvency check anyone can run against us.
/burns
[ Sample board ]Total destroyed
41,208,551 $EMBRR
Last 24h
892,004 $EMBRR
Funded by
revenue only, always
| Time | Spent · USD | Destroyed · $EMBRR | Transaction |
|---|---|---|---|
| 14:22 | 50.10 | 166,120 | 0x7a3f…c21e |
| 13:19 | 50.00 | 165,890 | 0x91cc…08b4 |
| 12:47 | 50.40 | 167,001 | 0x4d20…e7a9 |
/reserves
invariant: treasury balance ≥ sum of unspent credits
Illustration of the public burns page. It goes live with Phase 2; until then the real totals are zero and this board is labelled as a sample.
[ 09 ] The token · $EMBRR
It gets smaller every time somebody uses the product.
That sentence is the entire holder case. There are no distributions, nothing to stake and nothing to claim.
Without the token → with it
Full platform access
Same access. Nothing is gated behind the token
Pay the 10% fee
Fee discount, capped at 40%
New models on release day
New models 14 days early
No claim on anything
Still no claim on anything
The product works completely without the token. That is deliberate. A gateway that forces a token purchase before the first prompt has no users, and a burn with no users burns nothing.
Discount ladder · hybrid · capped at 40%
hold 0.25%
5%
hold 0.50%
10%
hold 1.00%
18%
hold 2.00%
25%
spend $100
5%
spend $500
10%
spend $2,000
15%
Up to 25% from holding plus up to 15% from rolling 30-day usage. Link a wallet by signing a message — no funds move, nothing is escrowed.
Why the cap is 40% and not 100%
A whale holds the threshold
pays no fee
and is also the heaviest user by definition
so the largest source of burn contributes zero
A 100% discount looks generous and quietly switches the machine off at exactly the accounts using the product most. Capping at 40% keeps every request generating a burn while still making the discount worth holding for.
10% × (1 − 0.40) = 6% — Effective fee never goes below 6%. It is a hard constraint, not a target.
What the token deliberately is not
Not a revenue share
holders receive no payments
Not staking
there is nothing to lock
Not governance theatre
we do not pretend a vote decides the model catalog
Not required to use
the product is open to all
Not a discount on cost
upstream model price is always paid in full; the discount only ever touches our margin
[ 10 ] Revenue, and the honest table
Small on purpose. This is a reseller, not a mint.
Four places money enters. Then the table we would rather you read before anything else on this page.
- 1Platform fee — 10%Charged on top of upstream model cost on every paid request. The primary and largest line.
- 2Developer API — 8%Lower rate, higher volume, no negotiation. A single integrated team can out-burn a hundred casual users.
- 3Prompt listing promotionAuthors pay to place a prompt above organic ranking. Clearly labelled. Ranking itself stays usage-based.
- 4Launchpad creator feeIncidental, not load-bearing. Excluded from every projection on this page on purpose.
Burn volume against real scale
[ A model, not a forecast ]$5,000 / mo upstream
$325 burned0.11% / mo · invisible
$25,000 / mo upstream
$1,625 burned0.54% / mo · faint
$100,000 / mo upstream
$6,500 burned2.17% / mo · a machine
$500,000 / mo upstream
$32,500 burned10.8% / mo · deflation
Arithmetic on hypothetical usage — assuming a 65% blended burn share of a 10% fee and a 300,000 USD market cap. Not a forecast, not a target, not a promise.
[ 11 ] Tokenomics
Fixed supply. One direction.
Contract facts
If the company acquires any position, it is bought on the open curve like everyone else and locked in a contract with a published unlock date. Not held in a wallet and described as locked. The difference between a claim and a fact is checkable in one call.
Launch sequence — why the token is not first
- Phase 1Gateway live, real usersNo token
- Phase 2Burn engine deployed, burns page carrying real transactionsNo token yet
- End of Phase 2Buyers can open the burns page and watch transactions land$EMBRR launches
The standard failure is launching the token on day one beside a thin product and a fee of zero, which produces a token whose only utility is a discount on nothing. By launching at Phase 2, the mechanism is evidence before it is a promise.
[ 12 ] Architecture
The client draws. It never decides.
Everything that touches a balance is computed on the server, and everything that touches the token is written to the chain where anyone can read it.
Browser / API client
draws the interface, computes nothing that touches a balance
server-authoritative core
Gateway
auth → quota → route → meter → bill
Upstream aggregator
hundreds of models, resale terms already held
Ledger
balances, usage, authorship — managed Postgres
Burn queue
accumulates to 50 USD or one hour
BurnVault
market buy → burn() → event, on chain 4663
Public burns page
one row per transaction, no login
What the client may never decide
Token counting
a client can under-report usage
Upstream unit cost
the basis of every fee calculation
Balance deduction
it is money
Discount tier
derived from holdings and spend, both verified server-side
Prompt attribution
decides who gets paid
Burn trigger
writes an on-chain transaction
BurnVault — reused, not written from scratch
The burn contract is an adaptation of a vault that already exists and is already tested: 21 unit tests and 3 mainnet-fork tests against the live launchpad, all passing. Only the trigger changes — it now fires on a settled revenue batch.
Not yet deployed to mainnet. Passing tests on a fork is not the same as being live, and this page will not blur the two.
A risk that does not reach us
Launchpad creator-fee routing on this chain can be reassigned by the factory owner through a timelock the current recipient cannot veto. Our burn engine is funded by AI revenue that never touches the launchpad, so that power exists and does not reach us.
Privacy — what is true
- Prompts and outputs are not stored after a request completes. Only metering rows survive: model, token counts, cost, timestamp.
- Nothing typed is used for training by us.
What is also true, and we will say it
- The request passes through our servers in plaintext in order to be metered and routed. We can see it while it is in flight. Anyone claiming otherwise while operating a gateway is overselling.
- The upstream provider has its own retention policy, which is theirs and not ours to promise away.
Stack
[ 13 ] Integrity
Every way to cheat it, and what happens.
Most of these defend themselves, because the attacker has to pay real upstream cost to try.
Attack
Defence
Free-tier farming
IP plus fingerprint limits, inexpensive models only, roughly 0.002 USD of exposure per visitor
Prompt-market spam
20 USD of lifetime spend required before a listing is allowed
Vote manipulation
There are no votes. Ranking is paid runs in a rolling window
Self-running for rank
Self-runs are excluded from ranking and from author earnings, and the attacker pays full upstream cost
Wash-burn for optics
Possible and pointless: the attacker funds our burn with their own money at a real loss
Withdrawal as laundering
Structurally impossible. Funds only ever return to the address that sent them
Runaway API spend
Hard per-key monthly caps set by the account, enforced server-side
[ 14 ] Roadmap
Product first. Token at the end of Phase 2.
Phase 0
Validation
days, blocking
We are here- □Upstream resale terms confirmed in writing
- □BurnVault adapted and re-run against a mainnet fork
- □Domain and handles secured
- □Model catalog and rate sheet pulled live
- □Legal read on prepaid credits and withdrawal
Nothing ships until the resale question has a written answer. It is the one failure that cannot be patched afterwards.
Phase 1
Gateway live
about one week
- □Text, code and image routing end to end
- □Metering and ledger with both invariants enforced
- □Free tier with the labelled simulation line
- □Cross-chain deposit, credits in seconds
- □Withdrawal to the depositing wallet, zero fee
- □Public reserves endpoint
No token in this phase. The product must stand on its own first.
Phase 2
Burn engine
three to four days
- □BurnVault deployed to chain 4663
- □Batching live at 50 USD or one hour
- □Public burns page with per-transaction rows
- □Burn receipt card and one-click share
- □Independent review of the vault before first fire
$EMBRR launches at the end of this phase, not before.
Phase 3
Prompt market
about one week
- □Listing flow behind the 20 USD spend gate
- □Automatic attribution and author settlement
- □Usage-based ranking, weekly reset
- □Video and audio added to the catalog
Phase 4
Scale
ongoing
- □Developer API at 8% with per-key caps
- □Command-line client
- □Hybrid discount tiers switched on
- □Team accounts with shared balances
- □Monthly burn-rate transparency report
[ 15 ] Growth, and what we measure
Five levers, one number.
Every lever is judged only by whether it moves monthly upstream spend.
- 01Burn receiptsEvery paying session produces a shareable artifact with an on-chain claim. Users distribute it because the post is about them.
- 02Prompt authorsAuthors take 30% and therefore market their own listings. Acquisition paid out of revenue, and only when it works.
- 03The withdrawal line"Unspent balance goes home" is one sentence, it is true, and an incumbent cannot copy it without rebuilding its float model.
- 04The proof pagesBurns and reserves are linkable arguments. In a category of unverifiable claims, a page that invites checking is the differentiator.
- 05Developer APISlowest to land, largest burn per account.
Watched weekly
- +Upstream spend per month — the only leading number
- +Paying accounts
- +Spend per paying account
- +Burn as a share of supply
- +Prompt runs by someone other than the author
- +Withdrawal rate — high is fine, it proves the promise
Deliberately ignored
Holder count. It moves with price, tells us nothing about the product, and chasing it leads directly to marketing the burn before the burn is real.
[ 16 ] Key decisions
Ten calls, and the reason for each.
Decision
Why
Upstream through an aggregator
Resale rights secured and a full catalog on day one; the thinner margin is the price
Platform fee 10%, API 8%
The aggregator already takes a cut; a higher rate would price us out of the comparison
Fee split 30 / 50 / 20
Authors paid enough to bother, burn is the largest single slice, operations kept lean
Discount capped at 40%, hybrid
A 100% cap would silence the burn at exactly the accounts that generate most of it
Credits withdrawable, depositing wallet only
The sharpest differentiator available, and the wallet constraint removes the laundering vector
Burn from revenue only
A treasury-funded burn is our own money in a circle; do it once and the metric means nothing
Buy on market, then burn
Burning inventory places no bid and only shrinks our own position
Free tier burns nothing
It costs us money and earns none; a burn there would manufacture a number
Token launches in Phase 2
Buyers should watch the mechanism work before they can buy exposure to it
No team allocation
Any position is bought on the open curve and locked in a contract, or it does not exist
[ 17 ] The honest take
Where we are strong, where we are not.
Six dimensions, scored without flattery. The weak ones are listed first on purpose.
First mover
Weak
We are not first. An incumbent already launched on this chain. We are second and should plan as the challenger.
Data advantage
Weak
We deliberately do not retain prompts, which forfeits the obvious data moat. Correct trade, but name it.
Network effects
Medium
The prompt market compounds: authors bring users, users make authoring worth doing. Real, but slow, and worth nothing until Phase 3.
Switching costs
Medium
Accrued discount tier, prompt library and author earnings — undercut by our own withdrawal policy. Also correct, also a real cost.
Technical
Medium
The vault is already written and tested against real mainnet state. The gateway itself is not hard to build and we should not pretend it is.
Brand and trust
Strong
Verifiable solvency, per-transaction burn records and a withdrawal promise we keep. In a category where every claim is unverifiable, being checkable is the position.
The actual moat — the refusal set
- ×never burn from treasury
- ×never charge a fee we cannot justify against cost
- ×never hold a token position outside a locked contract
- ×never claim privacy beyond what the architecture gives
- ×never market the burn before the number is material
- ×always return unspent balance
It is not the technology — a competent team clones the gateway in a fortnight. It is the refusal set. Each line is easy to copy on day one and painful to adopt on day four hundred, once float has been spent and a burn chart has been published. That asymmetry is the moat, and it only holds while every line stays true.
Risk
The single risk that outweighs the rest
The mechanism is sound and cheap to verify. The bet is not on the mechanism. The bet is that we can get several hundred people to move real AI spending onto this gateway. If that fails, the burn is arithmetic on a small number and the token is decoration — exactly the thing we built this to avoid. This is a distribution bet wearing a mechanism's clothing. Anyone evaluating it should evaluate it that way, and so should we.
Every prompt burns supply.
Other AI tokens sell you a discount card. This one gets smaller every time somebody uses the product.