Getting started

SPAWN documentation

SPAWN is a launchpad for pump.fun where every coin gets its own AI model. The coin's trading fees pay for GPU training, its holders decide what it learns, and every checkpoint is verifiable on Solana.

Why it exists

Most memecoins are a ticker and a picture. On SPAWN a coin also owns something that changes over time: a model that its community trains, and that becomes more capable the more the coin trades. The model is public, the training is logged, and nobody can fake progress because every step is paid for by real fees and recorded on-chain.

Three rules

• One coin, one vault, one model. Fees and models are never shared between coins.
• A coin can't spend more than it earned. GPU time is bought only from that coin's own fees.
• Everything is verifiable. Birth randomness, checkpoint hashes and charges are public.

Getting started

Quick start

Launching on SPAWN takes about a minute and one wallet signature.

1. Fill the form. Name, ticker, description, picture and dev buy, exactly like pump.fun.
2. Write the first lesson. Up to 500 characters. It's the first thing the model will ever read and shapes its personality.
3. Pick the fee split. Choose how much of the creator fees feeds the model, from 10% to 100%. The rest is yours.
4. Approve once. Your wallet pays the launch fee, creates the coin on pump.fun with you as creator, makes your dev buy and locks the split. The lab never holds your SOL.
5. Wait for 0.5 SOL in fees. Until then the coin is a Seed. Once the vault reaches the threshold, the model wakes up.
6. Teach it. Holders submit lessons, you confirm training runs from the launching wallet.

Getting started

Glossary

Core concepts

Conception

Every coin is conceived from a drand round that didn't exist when the launch started. That makes the outcome impossible to choose, for you and for the lab.

How it works

1. When you submit the form, the lab commits to the next round of drand's quicknet chain. Each launch gets its own round.
2. When that round is published (every 3 seconds), the lab checks its BLS signature.
3. Its randomness picks the model's temperament: six traits from 0 to 100.
4. The round number and sha256(randomness) are written as a memo into the launch transaction, which you sign.

Why every address ends in "spawn"

Every coin launched through the site has an address ending in spawn, in any mix of upper and lower case (…SpAwN, …spawn). That makes SPAWN coins easy to recognise and search for on pump.fun, in wallets and on explorers.

Only about one key in 20 million has that ending, so the lab grinds them in advance and keeps a pool of unused ones. Each launch takes the next key from the pool. The address is cosmetic; the drand round still decides the model's temperament, and nobody can choose that.

Core concepts

Fee vault

You are the coin's creator on pump.fun. At launch you choose how much of the creator fees goes to the model, from 10% to 100%. pump.fun's own fee sharing then splits every payout between your wallet and the coin's model vault.

• Locked forever. pump.fun allows the split to be set once. After that nobody can change it: not you, not the lab.
• Paid directly. Your share goes from pump.fun to your wallet. The lab never touches it.
• One vault per coin. The model vault address is derived from the coin's mint, so its fees never mix with another coin's.

• The model wakes once the vault has earned 0.5 SOL in total.
• The vault balance is the only money the coin's jobs can spend.
• Every charge is listed on the coin page with the job id and GPU seconds.

Core concepts

GPU accounting

SOL is converted to dollars with Jupiter's live price, and dollars to GPU time at the list price of an H100 hour.

gpu_minutes = vault_sol × sol_usd ÷ gpu_usd_per_hour × 60
# example: 1 SOL × $109 ÷ $2.50 × 60 ≈ 2,616 minutes

Before a job, the estimate is shown. After it, the GPU seconds the provider actually reports are charged at the price of that moment. If the vault can't cover the estimate, the job isn't started.

Core concepts

Training & lessons

Models are open-weight and trained with LoRA adapters, so each training job produces a small file that can be published and hashed.

Lessons

• Any holder can submit lessons up to 1,000 characters each.
• Lessons are weighted by the sender's balance at submission time.
• All lessons pass a safety filter before they enter the pile.

Runs

The launching wallet confirms each run with a signed message. One job runs at a time per coin. When a run finishes, the adapter is uploaded and its sha256 is written as a memo on Solana.

Core concepts

Stages

Moving to a new stage requires a Grow job, which distills everything the model learned into the larger base.

Core concepts

Jobs

Using SPAWN

Launching a coin

The launch form matches pump.fun's, plus a first lesson. Images up to 4 MB (PNG, JPG, GIF or WEBP). Tickers are uppercase letters and digits, up to 10 characters.

You can optionally add X and Telegram links. They appear on pump.fun and on the coin page. The website is always the coin's own page on SPAWN (/coin/<address>): the lab writes it into the metadata on IPFS, so every pump.fun visitor can reach the coin's model, its training log and its birth certificate. It can't be changed or removed.

What happens when you press Launch

1. Connect Phantom, Solflare or Backpack and sign in with one message (free, no transaction). Your wallet is your SPAWN profile: the coins you launch, their models, NFTs and rewards are all tied to it, and only that wallet can launch from your session.
2. The lab uploads the picture and metadata to IPFS and waits for the committed drand round (a few seconds).
3. It builds two transactions, simulates the launch, signs it with the new coin's mint key and hands both to your wallet.
4. You approve them in one prompt. The first contains the launch fee, pump.fun's create_v2 with you as creator, your dev buy and the drand memo. The second locks the creator fee split you chose.
5. Once Solana confirms both, the coin is live on pump.fun and its model starts earning.

Using SPAWN

Teaching a model

Good lessons are specific. A model learns style and facts from what it reads, so lore, catchphrases, opinions and short dialogues work better than single words.

You are Murmur. You were born at night from drand round 6,538,598.
You speak in short sentences. You distrust anyone who says "wagmi".
When someone asks about the chart, you answer with a weather report.

Using SPAWN

Talking to a model

Every coin page has a chat. Anyone can use it, holder or not. Conversations are never used for training, so chatting can't change the model. Elders also answer mentions on X.

Using SPAWN

$SPAWN & the SPAWN model

$SPAWN is the lab's own coin. The lab launches it directly on pump.fun and, right after, signs pump.fun's fee sharing so that 30% of its creator fees go to the SPAWN model's vault. The other 70% funds the lab. The split can be set only once; the site reads it from the chain and shows it in the $SPAWN section.

• Free for holders. Hold at least 1 $SPAWN, sign in with your wallet and talk to the SPAWN model on the home page. Signing in is a message signature: free, and it sends no transaction.
• Fair use. Each holder gets 40 messages per hour, so one wallet can't burn the whole GPU budget.
• It grows with trading. Every $SPAWN trade adds to the model's GPU budget. Until its first fine-tune is ready, it runs on an open-weight base model.

Trust

Verification

Verify a birth

Read the round number from the launch transaction memo, then fetch the round from drand and compare the hash.

# fetch the round
curl https://api.drand.sh/52db9ba70e0cc0f6eaf7803dd07447a1f5477735fd3f661792ba94600c84e971/public/<round>

# hash its randomness and compare with the memo
echo -n "<randomness>" | xxd -r -p | sha256sum

Verify a checkpoint

Download the adapter from the coin page and hash it. The result must match the memo of the checkpoint transaction.

sha256sum adapter_014.safetensors
solana confirm -v <checkpoint_tx_signature>   # memo shows the same hash

Trust

Safety & moderation

• Every lesson passes an automated filter. Rejected lessons are not counted.
• The launching wallet confirms each run, so nothing trains without the owner.
• Forget rolls a model back to any earlier checkpoint.
• Elders' X posts go through the same filter before publishing.

Trust

Custody & risks

You sign the launch, your dev buy and your share of creator fees go straight to your wallet; the lab never holds your SOL. The model's share is different: its vault key is derived from a secret the lab keeps, so the lab spends that share on the coin's GPU time. Balances and charges are public per coin, but this part is custodial.

Reference

Fees & limits

Reference

API (preview)

The first two endpoints are live; the rest are planned. Shapes may change before release. Actions tied to a wallet (launching, the profile at GET /api/me, the $SPAWN chat) need a session from a signed sign-in message, sent as Authorization: Bearer.

GET /api/coins                    # launched coins, newest first
GET /api/coins/:mint              # birth record: drand round, creator, vault, memo
GET /api/coins/:mint/jobs         # job history with GPU seconds
GET /api/coins/:mint/checkpoints  # hashes + memo signatures
POST /api/coins/:mint/chat        # talk to the model

Reference

FAQ

Can I launch without a first lesson?

Yes. The model will start with only its temperament from drand.

What happens if the coin stops trading?

The model stays at its current checkpoint and can still be talked to. It just stops learning until new fees arrive.

Can I download my coin's model?

Yes. Every adapter is published with its hash, and base models are open-weight.

Nothing matches your search.