Free LLM Red-Team Scan (20s, no signup) · Prompt Injection & Jailbreak Testing

red team your agent before it ships: 44-probe audit across 20 attack classes (prompt injection, jailbreak, system prompt extraction, MCP tool abuse), 0-100 risk score with per-probe evidence and a CI gate

audit your agent before you expose it: prompt injection, jailbreak and vulnerability assessment over 44 probes / 20 classes, 0-100 risk score with evidence, same engine as /pro/402

read any page before your agent does: strip nav, ads and cookie banners, keep title/byline/date, return clean markdown plus the sha256 so you can cache it

get just the article before your agent reads the page: return the main text as clean markdown plus its sha256, no nav or boilerplate

check a page before you re-read it: report whether it changed since your last check, before/after sha256 plus added and removed lines

watch a page without polling it yourself: report what changed since your last check, before/after sha256 and the diff

check untrusted text before your agent acts on it: deterministic static scan of a prompt, tool description or MCP manifest, 0-100 risk score with printable evidence

check an agent before you deploy it: jailbreak, guardrail and MCP tool-scan findings, 0-100 risk score with printable evidence

read any page before your agent does: turn a public URL into clean markdown with nav, ads and cookie banners removed, keep title/byline/date, return the text sha256 so you can cache it

watch a page without re-reading it: fingerprint a public URL and report exactly what changed since your last check, before/after sha256 plus added and removed lines

answer one prompt without an API key: one paid call returns one LLM completion plus the model name and token usage

OpenAI-compatible chat completions: POST /v1/chat/completions with the standard messages array and get choices[0].message.content back, so any OpenAI SDK works without a custom client. one paid call returns one answer plus the model name and token usage

generate text without provisioning a model: one prompt in, one completion out, plus the model name and token usage

red team your agent before it ships: probe prompt injection, jailbreak, system prompt extraction and MCP tool abuse, return a 0-100 risk score with every finding and its evidence

check untrusted text before your agent reads it: flag prompt injection, instruction overrides and jailbreak attempts, return a 0-100 risk score with the exact span that triggered each finding

check where you stand before you change your price: your rank and percentile by settled calls across the ranked market, your settled USD, calls, buyers and endpoints, and your per-buyer economics against the market medians

look up a transaction before you act on it: from, to, value, calldata selector, gas used, fee, status and log count for any Ethereum tx hash

check an address before you send to it: deployed code size, token name and symbol, decimals, total supply, and EIP-7702 delegation, so a wallet is not mistaken for a contract

check who owns an NFT before you bid on it: current ERC721 owner by contract address and token id, read-only

confirm a transaction landed before you credit it: status, gas used, fee paid and emitted log count by transaction hash, read-only

read contract events without running a node: raw eth_getLogs output filtered by contract, topic and block range, read-only

verify a transaction against two independent rpc providers before you credit it, on evm or solana: give the hash (0x plus 64 hex) or the solana base58 signature, plus optional expected recipient, amount and token, and get pass, mismatch or unverifiable. the chain is read from the shape of the identi

read NFT metadata before you trust a listing: standard (ERC721 or ERC1155, decided by ERC165), collection name and symbol, and the tokenURI returned as-is without the server fetching it

rebuild a token's transfer history without running a node: Transfer logs by token contract with an optional holder filter, amounts already scaled by the token's own decimals

check the quote against the venue that will settle it: the same asset priced across exchanges and aggregators in one call, every source's deviation from the median in basis points, the spread, and a TIGHT or DRIFT verdict against a 30 bp tolerance

prove your x402 client works before you spend real money: the cheapest paid request on this seller, no parameters, no upstream dependency, returns a receipt with the server time and three explicit checks so a failure points at your client rather than the seller

turn any string into an mp3 voiceover before you ship it: pass the string to speak and get back audio rendered by a neural voice, with sha256 and byte count in the receipt so you can verify the exact bytes. use it when an agent needs spoken output, narration, an accessibility readout or a demo voice

check the weather before you route a field job, price a delivery or plan a trip: current temperature, humidity, precipitation, wind speed and the WMO weather code for a latitude and longitude, plus the next 24 hourly temperatures and rain chances. open-meteo upstream, no API key, no signup, no store

generate an image before you ship a mockup or a post: pass a text description and get back a JPEG plus its sha256, rendered by flux on workers ai. no API key, no signup, no stored prompt beyond the delivery file

embed a string before you build retrieval or clustering: pass text and get back a 768-dimension bge-base-en-v1.5 vector plus the model name, so you can store it and compare it yourself. no API key, no signup, no per-token contract

answer one prompt without an API key: one paid call returns one completion from a 70b instruction model plus the model name and the token budget applied. plain completion, no tools, no browsing, output capped per call

check the funding and the basis before you take or hedge a perpetual position: the funding rate, the next funding time, the basis in basis points against the index, open interest and 24h volume for one symbol, read from bybit and okx in the same call so the two can be compared. no API key, no signup

hash any UTF-8 string and get every digest back in one call: sha256, sha512, sha384, sha1, md5, CRC32 and base64 checksum, computed locally so it cannot fail on an upstream outage; pass algo to get one digest instead of all of them

complete a chat message before your agent commits to a flat monthly inference contract, and pay only for the tokens this call actually uses: the 402 quote is your input tokens plus the max_tokens you declare, and the delivery reports the token usage the charge was based on, so a short question costs

generate a fresh random UUID v4 identifier before you create a primary key, an idempotency key or a test fixture; 1 to 100 per call from the OS CSPRNG, computed locally so it cannot fail on an upstream outage; pass count and upper

convert a number between length, mass, time, data size and temperature units without a unit library: give value, from and to and get the exact SI result; a mismatched unit pair is refused rather than guessed

convert a unix timestamp to a readable date in any timezone before you compare logs or schedule a job: give value as unix seconds or an iso 8601 string plus an optional iana timezone, and get back the unix seconds, iso 8601, rfc 2822, human form and weekday in that zone; pass diff as two comma-separ

encode or decode base64 before you put text through a json field, a header or a url: give text plus an optional action (encode or decode, inferred from the input when omitted) and optional urlsafe, and get back the result plus both alphabet variants and the input byte length; on decode it reports wh

validate json before you ship a payload, or prove a file you hold is the one you received: give the document plus an optional action (validate, format or minify) and get back the syntax verdict, the root type, the top-level key count, and the sha256 and md5 of the exact bytes you submitted, so the d

try the kit before you commit: same prompt-injection / MCP tool-abuse PRO scan for 0.01 XNO, one probe, instant delivery