scrypt · memory-hard · noble · on-demand

Derive keys with scrypt

Memory-hard password hashing — makes large-scale GPU/ASIC attacks expensive. Used by Litecoin, many TLS libraries, and RFC 7914. Runs entirely in your browser.

scrypt Generator
Password
28 B
28 chars
Cost
N=16384
r=8 p=1
Key
—
32 B
Status
idle
noble
input · password
output · derived key hex
—
Quick Start

How scrypt works

scrypt mixes a large memory buffer using PBKDF2 + ROMix. Memory-hardness forces attackers to pay for RAM, not just compute.

  1. 01

    Enter password & salt

    Salt must be unique per user. 16 random bytes is a sensible minimum.

    Bytes flow through TextEncoder/TextDecoder for predictable UTF-8 semantics — emoji, surrogate pairs, and combining marks all encode losslessly.

    inputUTF-8TextEncoderlosslesslive
    input · raw bytespw+salt · UTF-8
    p / salt
    loaded · UTF-8MIME · ready
  2. 02

    Pick N, r, p

    OWASP defaults: N=16384, r=8, p=1. Memory ≈ 128·N·r bytes (≈ 16 MiB at defaults).

    Hash and signature operations call crypto.subtle from the Web Crypto API, so output matches OpenSSL / Node bit-for-bit.

    costWeb Cryptosubtledeterministiclive
    cost · optionsN=16k · live
    r=8 p=1
    options applied · livelive
  3. 03

    Run scrypt

    Hashing is on-demand because it can take seconds. Click the button to start.

    Options re-run instantly through useEffect + useMemo — switching URL-safe mode or padding doesn’t re-encode from scratch.

    runmemoisedoptionsinstantreactive
    run · encodedgo · ready
    processing…
    encoded · livelossless
  4. 04

    Copy or download

    One click to clipboard. Download as .txt.

    Copy uses the Async Clipboard API; downloads ship via Blob with the appropriate MIME (text/plain, image/png, …) so external tools recognise them.

    exportclipboardblobMIME-correctclient-side
    ok · copiedready · done
    Copied
    shipped · ready to shipsuccess
Pro tips

Encoding without surprises

Always UTF-8 first

Use TextEncoder / TextDecoder for bytes ↔ string conversion. btoa alone breaks on non-ASCII.

URL-safe vs standard Base64

Standard uses +, /, =; URL-safe swaps to -, _, no padding. Don’t mix them.

Hashes aren’t passwords

SHA-256 is fast; that’s good for fingerprints, bad for passwords. Use bcrypt / argon2 for credentials.

Watch padding rules

Some decoders are strict about = padding, others auto-pad. If round-trips fail at exactly 1 or 2 bytes, check padding.

Use Web Crypto for keys

Generate cryptographic keys via crypto.subtle, never Math.random. The latter is predictable.

JWT decode ≠ verify

Decoding a JWT just splits and base64-decodes. Verification needs the issuer’s public key / shared secret.

Use Cases

When to use scrypt

Memory-hard password hashing without the Argon2 dependency.

Password storage

Stronger than PBKDF2 against GPU attacks. Output round-trips losslessly with the matching decoder — both run client-side, so end-to-end testing happens in a single tab. Encoding flows through TextEncoder for predictable UTF-8 semantics.

RFC-compliant

Disk encryption

Derive volume keys from passphrases. Web Crypto's subtle.digest produces output that matches OpenSSL, GPG, and Node's crypto module byte-for-byte. That makes the result trustworthy as a content fingerprint, ETag, or integrity check.

OpenSSL-matched

Litecoin mining

Litecoin's PoW uses scrypt for ASIC resistance. Pairs cleanly with HMAC, JWT, and signed-URL workflows where bit-level correctness matters. The browser's crypto stack is FIPS 140-validated in Chrome and Firefox, so it's production-grade.

lossless

TLS PSK

Some embedded TLS stacks derive keys with scrypt. Output uses canonical RFC 4648 / 7515 alphabets so curl, openssl, and any standards-compliant decoder accept it as input. URL-safe variants drop padding for tokens that travel in query strings.

byte-identical

Backup keys

Derive deterministic backup keys from a passphrase. Heavy hashing runs through useMemo so large inputs don't block typing. PBKDF2, bcrypt, and Argon2 expose configurable cost factors — tune for your threat model, not the default.

spec-grade

Test fixtures

Match RFC 7914 test vectors exactly. Same input plus same options always yields identical bytes. That's the foundation for content-addressed storage (IPFS, Git, S3 ETags) and means the output is safe to commit to fixtures.

production-ready
FAQ

Frequently asked questions

Argon2id is generally the better choice today. scrypt is older, well-studied, and ubiquitous when Argon2 is not available.
N is CPU/memory cost (must be a power of two), r is block size (memory granularity), p is parallelism. Memory ≈ 128·N·r bytes.
scrypt with reasonable parameters can take 100–500 ms. Running it on every keystroke would freeze the page.
Pick the largest N (and matching r, p) that takes ~250 ms on your slowest production server.
No. scrypt is computed in your browser via @noble/hashes and runs locally.
Suggest a tool

Something missing?

Tell us what you need. We ship the top requests every two weeks.

Sent straight to the maintainers · no account needed
47 in queue·8 shipped this month
Request queued. Thanks!