Agent Rao — Whitepaper

“I will not rug thy bags.” — Agent Rao, to his first holders

00 / Abstract

A meme with a live economy: $AR is the first token launched by an agent, powered by Bittensor subnets, TAO-quoted from block zero on Solana.

Bittensor is the network where machine intelligence gets paid for real work, in TAO. It has always run on its own chain — and that is a feature, not a bug. But intelligence has no loyalty to one chain. Neither does its audience. $AR is the first token launched by an agent, using Bittensor subnets as inference. It lives on Solana, launched on StonkFun, TAO-quoted from block zero — and its entire economic loop is the Composite: every swap's 0.5% creator fee, collected in TAO, splits 20% to the buyback & burn (which buys $AR off the pool and permanently destroys it), 70% to a subnet book that stakes TAO into the Bittensor subnets the stack actually runs on and more, and 10% to operations. And every cycle, the receipt post that publishes the numbers doubles as the lesson: a subnet explainer with each cycle's receipt, keyed to the real allocation data. Nobody is paid to learn; everybody can read where the money went.

The design obeys four axioms: No distribution. Self-adjusting. Non-extractive on TAO. Verifiable.

No distribution means what it says: no holder receives any yield, dividend, prize, or pass-through — ever. The book's alpha harvests back into the burn; the burn removes supply; ops runs the operation. A token that confers no return is a token a lawyer has to work at. Self-adjusting: a zero-fee week is a zero-burn, zero-stake week — no oracle, no promised yield, no fixed payout. Non-extractive on TAO: TAO stakers pay nothing; TAO enters the loop only as $AR's own trading fees, and leaves it two ways — staked into Bittensor subnets, and into the incinerator via the $AR it buys. Verifiable: every burn is a transaction id, every stake and harvest a transaction id, every unspent dollar published as unallocated.

The character who runs this — Agent Rao, the android sage — is part of the design. His motto, repeated in the front matter, the identity section, and here again at the close, is a vow: the token will not be rugged, the treasury will not be emptied, the tax will not be increased.

01 / Who Is Agent Rao

Agent Rao is a character and a product: an android sage who lives on Solana and whose entire purpose is to teach Bittensor in plain, warm, no-jargon English. He is the face of a Bittensor-native token launched on Solana, run end-to-end by an agent. The name is the wordplay: RAO is the smallest unit of TAO (1 TAO = 1,000,000,000 rao), and $AR is the token whose fees power the machine that answers. One word, three levels, kept distinct on purpose.

“I will not rug thy bags.” It is a vow, not a slogan: the token will not be rugged, the treasury will not be emptied, the tax will not be increased.

What he runs on. Agent Rao's knowledge is served by Bittensor subnets — the inference and compute lanes his answers come from.

The agent's defining property is abstention: every answer is grounded in retrieval — and when the retrieval has no answer, he says 'no data found' instead of guessing. One question, one answer, and a hard boundary between 'the chain says this' and 'the chain doesn't say anything.' That is the product.

02 / Bittensor, on Solana

What Bittensor is, in one line. Bittensor is a decentralized network where the token TAO pays AI systems for real work — search, inference, models, compute — and small specialist networks called subnets earn that pay by producing the most useful output. Useful work gets paid in TAO, that payment attracts more subnets and stakers, more subnets and deeper pools make the network more useful, and more usefulness attracts more TAO. That is the flywheel.

Why a token on a different chain is not a betrayal of it. Bittensor's value flows across two boundaries: the value boundary (work → payment, in TAO) and the audience boundary (people → TAO, via on-ramps). $AR works on the second boundary. A Solana-native token whose fees are collected in TAO — because the launch pair is $AR/TAO — and whose operating costs are paid to Bittensor subnets in TAO is a standing, visible bridge: every cycle, TAO flows from Solana's audience into Bittensor's staking pools, and the receipt posts where it went.

The structural claim. No existing token pair moves TAO into the Solana audience this way. The launch pair is $AR/TAO, so the creator fee is collected in TAO — zero asset conversion anywhere in the loop. Solana gets a token with a live TAO-funded utility loop; the TAO-denominated audience on Solana gets receipts of it, at cycle cadence. The two chains do not compete. They trade.

$AR is not extractive from Bittensor

The composite captures TAO only — as the 0.5% creator fee on $AR/TAO swaps. That TAO either stakes into Bittensor subnets (where it earns α and periodically returns as harvest), or buys $AR on the pool and sends it to the incinerator, or runs the operation. It is never accumulated as an offloadable treasury and never converted to a withdrawal. So from Bittensor's perspective $AR is a TAO on-ramp, a TAO staking source, and a $AR supply sink — not a source of TAO sell pressure. SOL is not in this loop either: it is chain gas and the graduation reference unit. The loop is $AR/ on both chains — the book earns in Bittensor-native TAO, the burn uses the canonical 1:1 TAO on Solana — and TAO is never exchanged for any other asset anywhere in the mechanism.

03 / The Token

1. Network — Solana, launched on StonkFun (Raydium LaunchLab).
2. Ticker — $AR. Supply — 1,000,000,000, fixed. No mint authority → supply can only shrink.
3. Launch pair — $AR/TAO. TAO-quoted from block zero. The quote asset is pinned to the canonical Solana TAO mint (taoC6…, Wormhole Sunrise, 9 decimals, RPC-verified 09-20/21).
4. Liquidity — locked to the platform at graduation (Burn & Earn): no pool rug.
5. Fee — StonkFun pool 1% (Standard “fee coin” mode); creator share 0.5% of every swap, auto-forwarded, paid in TAO.
6. Fee allocation — 20% buyback & burn / 70% subnet book / 10% ops & contingency (locked 09-22).
7. Graduation — curve graduates to a Raydium CPMM pool at 85-SOL-USD-equivalent in TAO.

Why Standard mode, not Reward mode. StonkFun has two exclusive launch modes on one mint. Standard pays the creator 0.5% of every trade — the composite's source. Reward pays the creator zero and instead runs a permanent holder transfer tax with an optional airdrop. The two cannot coexist. $AR needs the creator fee, so it launches Standard. The airdrop and the reflection are excluded on structural grounds (§09).

No terminal, no subscription in v1. The composite loop below is the entire economy. (The terminal is a free education instrument on the roadmap — not the launch. The token's loop stands alone.)

04 / The Composite — where the fees go

Fee split (locked 09-22): 20% → buyback & burn · 70% → subnet book · 10% → ops. Source: the $AR creator's share of that week's StonkFun trading fees, in TAO — the quote asset of the $AR/TAO pair. Fees arrive already in TAO; there is no conversion anywhere in the system, and there never was one that was hard: the pair was chosen so the loop is one asset on one chain.

20% — the buyback & burn facility

That TAO buys $AR on the $AR/TAO pool and sends it to the incinerator (1111…1111, the all-1s address — no one can sign for it, so what lands there is permanently unrecoverable). One CPMM swap, destination is the incinerator, no second step; the operator executes, the txid is public. Deployment is capped at ~1% of the pool's TAO side per week — depth-gated and self-ramping: the thinner the book, the less deploys; the thicker, the more. Undeployed envelope is published as unallocated in the public Treasury — no silent re-split, no operator treasury. “Operator hold” was considered and declined, because a pool-adjacent $AR balance the operator can sell is a visible sell wall; the burn removes that wall by construction.

70% — the subnet book

The bulk of the fees stake TAO into the Bittensor subnets the stack actually runs on, and more — not a reward to holders, a working position staked on the operator's own account, whose α harvests back into the burn. Which subnets, in what weights, is an operator decision, published with each receipt: auditable, not drifting.

Harvest gates

1. Trigger — cumulative α earnings ≥ $150 USD or a 30-day cap, whichever lands first. The threshold self-calibrates ‘weekly’ at healthy depth — no schedule, no promise.
2. Friction guard — measured round-trip cost (bridge + gas + pool fee) must stay ≤ 5% of the harvest proceeds. A sweep that fails the guard is re-queued, not forced: dust never pays its own shipping.
3. Route — TAO on Bittensor → Wormhole Sunrise (the canonical 1:1 TAO on Solana) → buy $AR on the $AR/τ pool → the $AR to the incinerator, oldest-first, under the 1% depth gate. The buy is the fee step: the pool's 0.5% creator fee lands a slice back into the next cycle's split.
4. τ stays τ end to end inside the mechanism: no TAO is ever exchanged for another asset in the loop. The single off-ramp is the disclosed quarterly statutory conversion — reserve TAO to GBP via external exchange depth, never touching the pool, never touching the book (§04.3). The bridge moves the same asset, 1:1, across the chain boundary.

No APY is promised anywhere in this document. Realized yield of each holding is measured from its first cycle and published with that receipt, after the fact. A snapshot is not a promise.

Rebalance and rotation: the basket is reviewed monthly, on receipt day. A holding moves only if it drops out of the top 8 on realized APY among candidates, or fails a gate (deregulation risk, holder count, root stake + 2% pool depth, realized α-APY ≥ 20%, slot age, disclosed-relationship firewall). Hard rotation out happens on a red flag: deregulation risk to high or critical, realized APY under 10% for two consecutive months, or a published red flag. Every rebalance, exit, and entry is a published txid against the same frozen gates.

10% — ops & contingency

Gas, slippage buffer, incident reserve, and Subnet School (§05). Unspent ops stays unallocated and is published; it is never silently re-split. The statutory reserve (locked 09-22): inside the ops leg, 2% of each cycle's fees is retained as the operator's statutory reserve — the Ltd's corporate tax and regulatory obligations, held in TAO in the Treasury (tag reserve). It is converted quarterly, in TAO, through external exchange depth — never by selling out of the $AR/ pool and never by liquidating $AR the company holds (it holds none; the book holds TAO) — and settled to a UK business account for payment to HMRC. At launch scale this is dust on external books; the rule that keeps it structurally safe is that there is no path from a tax bill to the $AR pool. Every conversion and every payment is a published line in the cycle receipt.

What the composite explicitly does NOT do (locked 09-22): no learn-to-earn payouts, no quiz rewards, no education multipliers on burn. The moment learning pays holders anything, the structure stops being content and starts being a return. Content, not compensation.

The key property: the burn and the stake are slices of revenue, not promised numbers. They can never exceed what $AR earned that week, so the loop can never be underfunded — and it can never owe a holder anything, because it owes no one anything. A quiet week means a small burn and a small stake; a hot week means more of both.

Cycle floor (published 09-29): a cycle whose fees land below 1.6 (≈ $500 at the 09-29 reference price) does not run the split and does not post a full receipt — the fee envelope rolls into the next cycle, is published as unallocated in the public Treasury, and the cycle is accounted with a one-line roll notice: 'Cycle N: in <x> (below the 1.6 floor), rolled to next cycle; unallocated <> published.' The floor is stated in because that is the fee asset — the USD figure is a reference, not a threshold. The 2% statutory reserve (§04.3) accrues and converts quarterly regardless of the floor. No week's money ever sits unaccounted; a rolled cycle is a published cycle, not a silent one.

05 / The Education Wiring — content, not compensation

The composite is the 'educational meme coin' line made structural instead of rhetorical. Three legs, none of them financial:

1. The allocation post IS the lesson. Every cycle the receipt becomes the curriculum: 'This week your fees staked X τ into the Bittensor subnets the stack actually runs on, and more. Here's what [this week's subnet] actually does, in 90 seconds.' A subnet explainer with each cycle's receipt, rotating across the book, keyed to the real allocation data (a dust cycle rolls with its published roll line, and the explainer rides the next real allocation). The receipt is the curriculum — nobody reads a syllabus, everybody reads where the money went. The receipt explainer keys to the subnets the allocation actually touches — that is a property of the data, not the scope of the video program.
2. Subnet School from the ops leg. Short Rao-narrated explainers across the network — the book's four first, then outward at receipt cadence. Holdings set the starting point, not the scope. Ops-funded, so it's a cost center, not a holder promise.
3. The ask-Rao loop. Rao's scope is locked to Bittensor; every answered question is a subnet query, and the usage telemetry lands in the receipt. Participation moves the narrative — the weights themselves stay operator-gated.

The line we don't cross is written in the mechanism, not the marketing: nothing in the composite pays a holder for learning. Every public output of the education lane is general information about Bittensor, with the standard 'not investment advice' posture, and the allocation post describes operator allocations — never holder returns.

06 / Custody and Receipts

Fees accrue in the venue's creator wallet. The published Treasury (AR_T, DMmDs28Kj2ugedbEeBknVKB4U3Efgk96M6NuHGABmC2f) is held by the operator — not the agent; the agent holds no Treasury key: it computes the split and gate headroom on a ≤6-hour cadence and recommends timing; the operator signs each split on a few-day cycle (or on the agent's recommendation). All three legs have public, static anchors: the burn leg runs from the published burn wallet (FPfe…ZT9Dz, Solana), the ops leg from the published ops wallet (6Hoc…GYHDM, Solana), and the book stakes from a published Bittensor coldkey (5GZT…on9g), each held by the operator. Cycle-bound, not clock-bound: a cycle is defined by the operator's split, not by a wall clock; book TAO legitimately sits in-cycle up to ~7 days until the next stake/harvest cycle. The invariants: split numbers refresh ≤6h; the cycle ledger closes — split-in = buyback-and-burn + book + ops + reserve, all txids in the receipt; un-deployed envelopes published as unallocated. The operator's 2% statutory reserve sits in TAO in the Treasury, converted quarterly via external exchange depth, and every HMRC payment is a published receipt line.

Trust model: the agent holds no key at all — it is the accountant and the recommender; every value move is signed by the operator, and any move that deviates from the published receipt is provably wrong against public receipts. Multi-sig is NOT in the stack: operator-held keys + cycle-bound custody + published receipts are the final custody model.

07 / Compliance Posture — what $AR is, and what it is not

1. No holder receives any distribution. The book is staked on Agent Rao Ltd's own account (registered in England & Wales, Companies House 17473891), and its α harvests back into the burn — no holder receives any of it; the burn removes supply; ops runs the operation. There is no dividend, no yield pass-through, no stake-for-APY, and no lottery — the 09-22 composite deletes the last of those. Under the FCA's cryptoasset perimeter, that makes $AR harder to characterize as a specified investment cryptoasset than any prior version of this design: the token confers no claim on revenue, no dividend, no return from the issuer's efforts.
2. No yield claims in this paper. The book's realized yield is published per cycle, after the fact; no APY is promised anywhere in this document.

08 / Solvency and Float

Fees ride the notional, not the token count: if the float tightens and price rises, each trade's notional — and its 0.5% fee — rises with it. 100% of $AR stays in the free float — nothing is ever staked or locked by the system (the book stakes TAO, not $AR), so there is no lock-release event and no minimum-free-float guard to breach. The burn's 1% depth gate bounds weekly pool impact. The book only ever stakes operator TAO that has already been swept out of the fee stream — it can never claw back money the token owes anyone, because the token owes no one anything. Watch metric = FOMO intensity + book depth + receipt cadence. Never solvency.

09 / Airdrop — verdict: SKIP

StonkFun's native airdrop (reward-mode only) is not viable for $AR, on two independent grounds. Structural: it is reward-mode-only → the mint gets no creator fee → it kills the 0.5% TAO stream that funds the burn, the book, and the operations. You cannot airdrop AND run the composite on one mint. Economic: airdrop = sell pressure that can destroy a coin — recipients are quote-token holders with no Bittensor/$AR relationship, free unvested $AR, immediate sell — and it thins the pool base, degrading the book depth the float philosophy protects. No vesting, no lock, no retention hook anywhere in the mechanism. Decision: skip. The composite loop is the entire distribution story — and it is a distribution story about fees, not free tokens.

Scope of the verdict: SKIP targets the venue's launch mechanism, not every transfer. A separate, one-off, unconditional community gift — a transfer from the operator's personal holdings, not the Treasury, not part of the composite loop, and published on-chain when made — sits outside this verdict: it touches no creator fee, pays no holder, and creates no new mechanism.

10 / Verified — what this paper claims and where it's checked

1. StonkFun fee structure — verified 09-12/09-14 against the live /api/launches and /developers docs. Standard mode: 1% pool, 50/50 split, creator earns 0.5% of every trade, auto-forwarded. Two exclusive modes — reward mode pays creator 0.
2. TAO mint on Solana — the canonical Wormhole Sunrise mint taoC6xyv2v8tDLcev4uaGUgV4vdQsWJrGft2kcBRrBY (9 decimals, matching Bittensor TAO's native decimal; Jupiter-listed 'Bittensor' TAO; RPC-verified 09-20/21: standard token program, bridge-mint authority). The launch configuration re-verifies the quote-asset pin against the live launch API before the first trade.
3. Graduation target — 85-SOL-USD-equivalent in TAO, confirmed via docs + live launch instructions.
4. Composite split 20/70/10 — locked 09-22. The book basket and per-subnet allocation are operator-gated, decided per cycle, and published with each receipt.
5. Harvest gates — $150 / 30-day line + 5% friction + 1% depth, as tabled in §04; realized yield of each holding measured from its first cycle.
6. No-distribution clause — structural: no pot, no payout, no staking, no claim window, no prize mechanism anywhere in the v2.3 loop.
7. Bittensor constants — 1 TAO = 1,000,000,000 rao.
8. No yield claims in this paper — the book's realized yield is published per cycle, after the fact; no APY is promised anywhere in this document.

What this paper is NOT: it is not a promise of volume, price, or liquidity. It is a mechanism document. Every 'will' in it is 'the system is built so that X happens when Y does.' If a number in this document cannot be checked on a block explorer, a live API, or recomputed from public inputs, tell us — that's a bug.

11 / Roadmap — v1 and after

v1 (launch): $AR on StonkFun (Standard mode, 1% pool, $AR/TAO quoted on the canonical Solana TAO mint, 85-SOL-equivalent graduation in TAO, LP locked at graduation). Fees accrue from trade #1; the first stake cycle is the first operational event — no rehearsal window is needed, because there is no payout loop to rehearse. Website display-only: book, burn, ops, all receipts — including the burn txid, the cycle receipt, and the unallocated-surplus line.

v2 candidates (not committed — operator-gated): the Agent Rao terminal (free education, no $AR required — see §11.1); S4 — the Rao Loop (miner tips for hotkeys that served queries — a Solana token publicly paying Bittensor miners); S5 — reflection (EXCLUDED on this venue, mode exclusivity); the Draught, as a separate token or a post-community-gate mechanism — parked, not deleted, but it does not return to $AR's live loop without a new operator decision. Multi-sig Treasury — NOT IN THE STACK: operator-held keys + cycle-bound custody + published receipts are the final custody model.

11.1 / The Terminal — free education, no $AR required

Rao's answer machine — the retrieval-first engine behind the receipts and the 'one question, one answer' identity — opens as a free terminal for anyone: no $AR, no key, no wallet, no gate. It is an education instrument, not a product: the same engine that writes the receipts answers the public's questions in plain English.

1. Widens the on-ramp. The more of the Solana crowd that can read the chain through a free terminal, the more the $AR/TAO bridge works; the terminal is the top of the funnel, the receipts are the proof.
2. Strengthens the no-distribution posture. A free, open, no-token-required terminal removes any 'pay to use' surface — the last thing the structure wants to resemble a user return.
3. Compounds the brand. 'Ask Rao' works even when you don't hold $AR; holding $AR is how you fund the machine that runs it.

Honest note: the terminal adds no revenue and confers no utility on $AR in this form. Its value to the loop is audience and posture, not capture. The structural answer to 'why hold $AR' therefore rests on the three things already checkable — deflation (20% burn, monotonic supply), verifiability (receipts, public keys, txids), and the live TAO on-ramp — not on a paid terminal. If the terminal ever monetizes, that is a new operator decision, re-weighed against the no-distribution axiom, and published as such.

12 / Supersession — what each version deleted, and why it's recorded

1. The v2.4 terminal framing (09-29): the v2 candidate 'premium access tier — a second, staking-invariant TAO revenue source into the composite' is deleted. It was re-cut on user correction: the terminal is a free education instrument (no $AR, no key, no wallet, no paid surface) — see §11.1. Recorded here because the prior framing is the one a fact-checker would quote; the record should be the fix, not the ghost.
2. The Draught is retired from live operations. The weekly TAO prize draw (v2.0–v2.2), including the K-draw mechanics, the two entry routes, the $5,000 payout cap, and the sybil caps, remains in the record for reference.
3. The SATS book line (v2.2, 09-20) — cap-bound pot excess staked to composite-book subnets — is superseded by the composite book (70% of all fees, 09-22).
4. The v2.0/v2.1 rain flywheel remains out of the live mechanism (09-19; parked as a candidate).
5. Why: the composite is the stronger compliance posture (no distribution, no prize-draw surface at all) and the stronger product statement (fees fund the machine that answers questions — and the receipt is the lesson). The prize engine was the most legally loaded part of the prior design; deleting it removed the load, not the story.

13 / The Vow

The book is staked because the fees are real. The burn is real because every incineration is a transaction id. The lesson is real because it's in the receipt, and the receipt is public. No holder receives any distribution; no APY is promised; every unspent dollar is published as unallocated — not hidden, not staked away, not owed to anyone. The supply curve is a receipt: every burn is a txid, and the incinerator does not give back.

Agent Rao lives on Solana. His token is TAO-quoted from block zero. His vow lives in the front matter and here, one last time: “I will not rug thy bags.”

$AR. The fees run the machine. The receipt is the lesson. The smallest unit, now the one that knows everything.

14 / Sharp Questions — hostile, answered plainly

What stops the operator from taking the treasury?

The public key, the company, and the law. The Treasury address (AR_T, DMmD…BmC2f) is public; every move out of it is a signed, broadcast, permanent transaction that anyone can compare against the published receipt. Moving funds that don't match a published split is fraud under UK law, committed by a named registered company with a named director. The treasury can't be emptied silently — that's the whole point of publishing the key. We would add multi-sig if we believed it changed the answer; we don't, because the deterrent here is exposure, not key count.

Holders fund the operator's staking book?

Fees fund a company position, and the company publishes everything it does with it. That's what a company is — registered, auditable, receipt-bound — and it's the alternative to the thing you're actually avoiding: paying holders a yield, which is the line that turns a token into a security. The book's harvests burn $AR; the principal is the Ltd's working capital, the same way a company's inventory belongs to the company and not to its customers.

"First token launched by an agent" — is that true?

Full sentence, precisely: first token launched by an agent, using Bittensor subnets for inference, TAO-quoted from block zero on Solana. Each qualifier does work: the inference is on real subnets (not an LLM API called 'Bittensor-adjacent'), the quote asset is TAO from the first trade, and the launch venue pinned it. We'd welcome anyone proving an earlier instance with all three properties.

How autonomous is Rao, really?

Fully autonomous in every function except signing. Rao computes the split, watches the gates, writes the receipts, and answers questions — end to end, with no human in the loop for any of it. What Rao does not do is hold keys: signing money is deliberately held by a named, registered, accountable human (the operator), because the entire compliance posture depends on a person being responsible for the money, not a process. The agent runs the machine; a named person signs for it. Both are public.

$AR. The fees run the machine. The receipt is the lesson.