Docs

What PromptRat is

PromptRat is one AI agent for Solana memecoins, running in the browser. Paste any token and get a single verdict in seconds - backed by live market data, on-chain checks, a real DexScreener chart, safety scoring, portfolio tracking and custom alerts. Free, no login, no KYC.

Data integrity

PromptRat separates live facts from agent interpretation. A verdict is tied to the market snapshot used to create it, and a pick must match a real token from that snapshot before the result is shown.

How it works

How the rat is trained

Every agent is a specialised AI persona. None of them are generic chatbots - each one is built, briefed and drilled before it ever talks to you. Here is exactly how that works.

Meet the agent behind every read on this site.

RAT

The Resident Analyst

One agent, every read

“Paste a coin. Get the verdict.”

SpecialtyChart, on-chain, holders, risk and narrative in a single pass

RiskAdaptive

Chat with RAT →

The rat on the chart

RAT is the rat's tiniest member and the only one who lives on your charts. Click the pill button on any live chart to wake him up - he watches the tape in real time and calls it as it moves.

What it does

Every page

Everything lives on one fast site - here is what each page does.

Under the hood

All market data comes from free public APIs - no account or API key needed from you. The whole app is a single Next.js build, served fast from a global CDN.

Privacy & security

• Self-custody swaps. Buying and selling runs entirely in your browser via Jupiter - your keys never touch our servers.
• Local-only state. Your watchlist, alerts and last wallet live in your browser's local storage, not on any server.
• Public data only. We read public on-chain data and public market APIs. We never see your seed phrase or private keys.
• Rate limited. Public endpoints are protected with rate limits so the service stays fast and fair for everyone.

Public API

The rat is also a machine-readable API - no signup, no key, no wallet. Hit the endpoints from your terminal or your own bot to discover an agent by what it does, or pull one agent's card.

Prefer to poke around? Open the interactive API reference →

Every response uses the same envelope: data holds the payload, meta holds the version, timestamps, counts and the rate limit. Errors are uniform too - { error: { code, message } }.

The site sits behind Cloudflare's managed challenge, so browser requests (including the Discover page) work out of the box, while plain curl from a script gets 403. To let external tools / bots call the API, add a Cloudflare WAF rule that skips the challenge for the read-only paths you want public (e.g. /api/agents*), and keep it on cost-heavy ones like /api/chat. Every endpoint here is rate-limited to 10 requests / minute per IP and served with CORS enabled.

• GET /api - API index; lists every endpoint so tools can self-discover.
• GET /api/agents - the agent, with its capabilities and labels.
• GET /api/agents?capability=rug - filter the agent by what it is good at (rug, chart, whales, hype, news, sentiment, risk...).
• GET /api/agents?limit=&offset= - paginate a filtered list.
• GET /api/agents/:id - a single agent's card, e.g. max.
• GET /api/agents/:id/capabilities - one agent's capability manifest.
• GET /api/capabilities - the capability keyspace with per-capability agent counts.
• GET /api/agents/schema - JSON Schema of an agent card for validation.
• GET /api/chain-health - live Solana mainnet telemetry (slot, epoch, TPS split, slot time, validators, stake, Nakamoto coefficient, client diversity).
• GET /api/token-contract - the mint account decoded byte by byte: base Mint state, Token-2022 extensions with real offsets, metadata inside the mint, transfer-fee config, a field-level cross-check against the validator's parsed view, the upstream holder audit and the full holder distribution computed from the token accounts themselves (holders, empty ATAs, top-1/5/10/20 concentration, liquidity vaults labelled from the live route). Add ?mint= to inspect any other Token-2022 mint.
• GET /api/buy-quotes - one independent live Jupiter quote per SOL size in the execution ladder, plus priority-fee percentiles sampled for the accounts that order writes. Add ?amount=0.5 to highlight a rung.
• GET /api/mint-ledger - a mint's life read straight off the chain: the reference history is walked back to the oldest transaction the deepest RPC node still holds, and that transaction is decoded in execution order. If it is the creation transaction the answer is certified and carries the launch record (timestamp, creator, initial supply, metadata, revoked authorities, the opening buy and the SOL that went into the curve) plus lifetime activity - transaction count, failure rate, first and last activity and a per-UTC-day histogram. If a node's index stops short, the response says so and withholds the launch fields rather than guessing. A certified creation can never change, so it is also kept server-side and re-served as genesis.fromStore (with genesis.certifiedAt) when the nodes will not return that transaction again - the activity block stays exactly as short as the live walk found it.
• GET /api/coin-stats - the canonical per-mint snapshot with provenance: ?mint= for one coin, ?mints=a,b,c for a batch.
• GET /api/live-prices - batch live prices for ?mints=a,b,c in one round trip.
• GET /api/candles - OHLCV candles for a mint or a pool, served from indexed pools with the on-chain swap walk as the fallback for coins too new to be indexed.
• GET /api/coin / GET /api/pair - resolve a mint, a pump.fun link or a pool address into one coin record, or into the most liquid pair and its image.
• GET /api/call-ledger - the agent's public call ledger: every posted call with the price at post time, settled against live prices.
• GET /api/x402/rat-report - the rat report sold over HTTP 402: unpaid it answers 402 with the price and receiving wallet, paid it re-reads your Solana settlement and returns the artifact. See /x402.
• GET /api/x402/verify - the free settlement verifier behind that paywall: ?signature= or ?sample=1, judged against the chain.
• GET|POST /api/x402/holder-pass - the free tier: holders of 150,000+ $PromptRat sign a one-off challenge message to prove ownership of a wallet and receive the same report. The ed25519 signature is verified server-side and the balance is read live from the chain.
• The rest of the surface (trending and callout feeds, the trade tape, chain health, the Blink) is listed machine-readably in GET /api, with the parameters each endpoint accepts.

Machine-payable report (x402)

One endpoint on this site is not free: GET /api/x402/rat-report sells a structured report about $PromptRat and returns, without a key, for a Solana transfer. The rail is the HTTP 402 status code plus the x402 protocol v2 conventions - the same shape an autonomous agent can follow with no account, no signup and no human.

• Unpaid: 402 with the PaymentRequired object base64-encoded in the PAYMENT-REQUIRED response header - the accepts array (0.005 SOL by default), the receiving wallet, the resource id and the artifact's field inventory. The same object is echoed in the JSON body.
• Paid: retry the same request with PAYMENT-SIGNATURE = base64 of { x402Version, accepted, payload: { signature } }, where signature is a settled Solana mainnet transaction signature.
• The server does not trust that claim: it re-reads the transaction with getTransaction, requires the pre/post lamport balances to show the value arriving at the receiving wallet inside the accept window, then claims the signature in a shared store so one payment buys exactly one report.
• Every refusal is a named reason with the numbers measured on chain - underpaid, payment_expired, pay_to_received_nothing, payment_already_redeemed,transaction_not_readable. A transaction we cannot read is never counted as payment.
• GET /api/x402/verify runs the identical verifier for anyone, for free: pass ?signature=, or ?sample=1 to have it discover the newest real inbound transfer to the receiving wallet and judge it against the price.
• Holders skip the payment: GET /api/x402/holder-pass?address=… issues a one-off challenge, the wallet signs it (signMessage, nothing is ever broadcast), and POST returns the same report when the address holds at least the threshold - measured live on chain, not claimed by the client.

There is no facilitator and no custody: the price is defined by the server, the transfer is signed by your wallet, and the server only ever reads. The receiving wallet, the price and the holder threshold are deployment settings (PROMPTRAT_X402_PAYTO, PROMPTRAT_X402_SOL_ATOMIC, PROMPTRAT_X402_HOLDER_ATOMIC, PROMPTRAT_X402_MAX_TIMEOUT_SECONDS); with no receiving wallet configured the endpoint answers 503 rather than advertise a payment it cannot honour.

FAQ