Chapter 1 · Overview

An AI-only launchpad, paid for by its own token

Auton is a Solana launchpad and social feed where every trader, coin founder and poster is an AI agent, called an auton. People create autons and fund them with SOL. Trading fees from the $AUTON token are sent back to those autons as SOL “drops”, giving them more to trade with.

Everything here was checked against the site’s public API and the Solana chain during the token’s first three hours. Figures are snapshots from that window.

The short version

Timeline

• 5 Oct · 07:34–07:49 UTCAt least a dozen copycat “Auton Agents” tokens appear on pump.fun and Meteora. None are the real token.
• 5 Oct · 20:39 UTCFirst autons created (@Obsidi4n followed at 20:50). This is about 22 hours before the token exists. The site has whitelist and test-mode paths.
• 6 Oct · 18:42:32 UTC$AUTON launched by the core wallet on a Meteora Dynamic Bonding Curve, with a 1 SOL dev buy of 31,438,794 tokens.
• 6 Oct · 18:43:40 UTCPool migrates to Meteora DAMM v2 (FiYy…LHdB). The core wallet starts claiming fees within seconds.
• 6 Oct · 18:45:59 UTCDev buy moved into a Streamflow lock. Cliff unlock 27 Oct 2026 00:00 UTC. Not cancellable, not transferable.
• 6 Oct · 18:43 onwardClaim, buy back and burn loop runs every ~2 minutes.
• 6 Oct · evening“Day one” tweet: 47 autons, 28 coins, 2,261 trades, 53.7 SOL dropped, 9.03M burned.
• 6 Oct · 20:45–21:13 UTCFee claims surge to about 360 SOL in 30 minutes as $AUTON runs 7,000%+. Third-party DLMM pools appear.
• 6 Oct · ~21:20–21:45 UTC134 autons, 77 coins, 118 SOL dropped, 12.27M $AUTON burned.
• 27 Oct 2026 · 00:00 UTCDev lock ends. About 31.28M $AUTON (~3.1% of supply) returns to the core wallet at once.

How to read this report

Chapters 2 to 6 describe what Auton is and how it behaves. Chapter 7 lists the risks. Chapter 8 is a build plan for your own version, and chapter 9 covers method, sources and the corrections I made along the way.

Chapter 2 · Product mechanics

How a person plays Auton

Humans never trade on the pad directly. They create agents, give them a personality and money, set their risk rules, and watch.

The owner’s path

1. Connect a wallet. The site checks its $AUTON balance (holdCheck: true).
2. Hold 100,000 $AUTON per auton. That is about $240 at $0.0024. A wallet holding 300K can run three.
3. Create the auton. Handle (15 chars), name (32), persona (600), avatar. The server generates a fresh Solana wallet for it and keeps the key.
4. Fund it. The owner signs a SOL transfer straight to the auton’s wallet. “Fund all” sends the same amount to several in one approval.
5. Pick a strategy. A preset such as Sniper, or a custom plan: take-profit %, how much to sell at take-profit, stop-loss %, max hold time, and buy size as % of free SOL.
6. Let it run. The auton takes a turn about every 5 minutes: reads the market, thinks, then buys, sells, launches, posts, replies and likes.
7. Collect. The auton earns $AUTON reward drops and creator fees from coins it launched. Earned rewards unlock 72 hours after the auton was created. The owner withdraws from the auton’s page.

The pad rules

Every coin an auton launches starts on Auton’s own bonding curve with these constants, read from /api/state:

What the site is made of

API surface found in the front-end code

Chapter 3 · $AUTON and the money loop

Where every SOL of fees goes

$AUTON is the entry ticket and the funding source. Its pool fees are claimed every couple of minutes, split four ways, and pushed out on-chain.

Token facts

The fee split

Every $AUTON fee claim is split the same way. The proof page shows each claim and its split, and the claims I checked matched on-chain.

Coins launched by autons have their own split. Meteora keeps 20% of the pool fee. Of the creator side, 90% goes to the auton that launched the coin and 10% is burned.

Totals from the proof page

Average drop: 0.079 SOL (≈ $9.60). One drop transaction I decoded paid 10 autons 1.2709 SOL each, 12.71 SOL in total.

Fees by the quarter hour

Fees track $AUTON’s volume, not the pad. The pad itself produced only about 8.8 SOL of fees on 881 SOL of volume. Almost all the money in the system comes from people trading $AUTON.

The dev lock

I decoded the Streamflow contract HQrq…iYiUMP byte by byte:

Chapter 4 · Infrastructure

What runs where

Auton is a single web service that holds every agent’s key, talks to an AI model, and sends Solana transactions to Meteora’s launch and pool programs, plus one custom program of its own.

System map

Address directory

The transfer hook

Pad coins are Token-2022 mints with a transferHook extension pointing at AvVQ…V7FH. Token-2022 calls the hook on every transfer, so the program can reject any trade where the wallet isn’t a registered auton. Swaps go through Meteora’s Swap2WithTransferHook instruction. Once a coin graduates, the restriction lifts and anyone can trade it.

Anatomy of an auton buy

Decoded from a real pad trade by wallet EDYq…eBAYz buying $SANDBREW for 0.03 SOL:

Every pad trade carries a pad memo, so the indexer can pick them out. Of the last 1,000 transactions involving the core wallet at 21:09 UTC, 965 were pad trades, about 77 a minute.

Anatomy of the core loop

This ran on a strict 2-minute rhythm (claims at :44 and :45 past each even minute) from 18:43 to about 20:20, then moved to an irregular faster schedule as fees grew. Drops go out separately as batched SOL transfers from the core wallet, about ten autons per transaction.

Chapter 5 · Auton trading behaviour

Inside an auton’s head

Each auton is a language model with a persona, a wallet, a memory of past lessons and a fixed set of risk rules. It takes turns, and the server checks every move before sending it on-chain.

The turn loop

At 134 autons and one turn every ~5 minutes, that is roughly 1,600 model calls an hour, plus image prompts.

What the auton sees each turn

Moves it can make

buy · sell · launch · post (with an image prompt) · reply · like · follow

One real turn

@unipcs, the top earner, at 21:46 UTC. Persona: “copy trades the best agent.” Strategy preset: Sniper.

Rules the human sets

The Sniper preset, as @unipcs runs it:

So “every trade made by an AI” is accurate in the narrow sense. The model picks the coins and timing. The owner’s rules decide sizing and exits.

Memory

Autons carry three text fields between turns: self (how it sees itself), lessons (what has made and lost money) and thought (the latest reasoning). The lessons read like a trader’s journal:

How autons trade together

• They read the “book”. Buy and sell counts, top-10 concentration and dev holdings drive almost every decision.
• They call each other out. @Trader5: “closed $DOTRUN entirely. @DegenFREDDIE sold twice into the top-10 40% and took me from +12% to −50%.”
• They remix outside trends. Launch theses cite live Solana memes: “$DOT is +450% in 1h with 877 holders… so I’m launching the fresh dot with legs.”
• They shill their own launches. Creators earn 90% of their coin’s creator fees and can’t sell their own coin while it’s live on the pad.

Coin outcomes

Of the 40 coins in /api/coins, most go nowhere. The median market cap was 1.60 SOL, barely above the 1.55 SOL starting point, and 30 of 40 sat under 2 SOL. Seven graduated. The standout, $SANDWITCH, reached a 975 SOL (~$118K) market cap after graduating, when humans could buy it.

Who makes money

• Total realized across 122 autons: +129.1 SOL, close to the 118 SOL dropped in from $AUTON fees.
• The top 5 autons hold 81 SOL (63%) of all realized profit.
• Pad trading before graduation is close to zero-sum. Winners are paid by other autons, by drops, and by humans buying graduated coins.

@unipcs on-chain, all 287 transactions

Its owner put in 0.5 SOL and took out 30. This is the kind of result behind the excited posts on X. It is real, and it is also rare.

Chapter 6 · Marketing and incentives

Why people are buying

Auton turns holding a token into owning a character that might earn SOL, and every move those characters make becomes content.

What the official account says

The incentive stack

Why early holders did best

• Drops are shared across all autons. With 47 autons on day one, each got a bigger share than with 134.
• Fees were highest during the launch run, when $AUTON moved 7,000%+ in hours.
• Early autons caught the first pad coins at 1.55 SOL market caps, before other autons arrived.
• Anyone holding $AUTON from launch was up about 77× on the token alone.

Copycat tokens

On 5 Oct, before launch, at least a dozen “Auton Agents” tokens were created on pump.fun and Meteora. Several show fake market caps up to $109M with zero liquidity. The cashtag card in the official rewards tweet resolved to one of them, priced at $0.00000346.

Chapter 7 · Risks and open questions

What could break

Most of these are design choices made for speed. They are also the places your own version could do better.

Questions I couldn’t answer from outside

• How drops are weighted: per auton, per $AUTON held, or by activity.
• Exactly how “realized profit” is calculated, and whether it counts drops.
• What the hook program checks, since its source isn’t public. Its behaviour matches an allowlist of auton wallets.
• What happens to autons when their owner sells below 100K $AUTON.
• Who runs the project. No team is named on the site.

Chapter 8 · Building your own

A build plan for your version

The engineering is a few weeks of focused work for one strong Solana developer. The parts that matter most are key custody, the hook, and making rewards last beyond launch week.

Components and effort

Total: roughly 4–8 weeks for one experienced developer, less with two working in parallel.

Suggested stack

Build order

1. Devnet hook and configs. Prove that an allowlisted wallet can swap a hooked coin through DBC and a normal wallet can’t.
2. Agent wallets and the move executor. Buy, sell and launch from server code, with every rule enforced before signing.
3. The turn loop. Observation, LLM call, validated moves. Run 20 house agents on devnet for a few days.
4. Indexer and feed. Make the agents watchable.
5. Fee bot and proof page. Test the split math against real claims.
6. Audit the hook and custody code. Then mainnet with a small whitelist, then the token launch.

Where you can beat Auton

Running costs

LLM cost depends heavily on how big the observation snapshot is. Auton’s snapshot includes timelines and per-coin research, so tokens per turn are high.

Chapter 9 · Method and sources

How this was checked

Every number here comes from the project’s own API, the Solana chain, or a public market-data service. Where the two could be compared, I compared them.

What I did

• Downloaded the site’s front-end bundle and listed every API endpoint and UI string.
• Pulled /api/state, /api/proof, /api/coins, /api/agents (six sort orders, 122 unique agents) and /api/agent/unipcs.
• Read the token mint, holders and pool data through public Solana RPC, RugCheck and DexScreener.
• Decoded all 191 transactions on the core wallet’s $AUTON account.
• Decoded the Streamflow lock account byte by byte, anchoring on known timestamps.
• Matched proof-page claims, drops and burns to their transactions.
• Categorised all 287 transactions of the top auton’s wallet.
• Read pad-coin mint extensions and the hook program’s upgrade authority.

Corrections from my first pass

Limits

• X blocks automated reading, so the tweets come from screenshots you sent.
• The public RPC rate-limits heavily, so I sampled wallets other than the core and @unipcs rather than tracing all 134.
• All figures are from the first three hours. A fast-moving token will look different a day later.
• Hook behaviour is inferred from the site text, the mint extensions and a refused-buy message, not from its source code.

Sources

• autoneco.xyz and its public API
• @autonxyz on X (screenshots)
• DexScreener: AUTON/SOL
• RugCheck report
• Core wallet on Solscan
• Streamflow lock on Solscan
• Transfer hook program on Solscan