juniper9-crypt

[!IMPORTANT] This repository is read-only and no longer developed. Development has moved to network-secret, which brings juniper9-crypt, juniper8-crypt, and the Nokia SR OS custom-hash cipher together in a single package with a single CLI.

Migrating is an import path change and nothing more: pip install network-secret, then from network_secret import juniper9. The encrypt(), decrypt(), and check() signatures are identical.

Encrypt and decrypt Juniper $9$ reversible passwords, from the command line or Python.

The $9$ algorithm is a proprietary Juniper substitution cipher. It is keyless and device-independent: a password encrypted on one Juniper device can be decrypted on any other, with no node-specific secret involved. The algorithm and its character set are publicly documented.

Prefer a browser? Encode and decode $9$ (and $8$, and Nokia SR OS custom-hash) at network-secret.pages.dev. It runs the same algorithm fully client-side - nothing you type is ever sent to a server.

$9$ is a substitution cipher, not real cryptography. Treat it as obfuscation, not protection. Anyone with this library (or the source of any Juniper device) can recover the plaintext.

Run without installing

If you have uv installed, uvx runs the CLI without installing anything:

uvx juniper9-crypt --decrypt '$9$FNkC3/t1IcevLuOWx'

Don't have uv yet? Get it. It's the best thing to happen to Python tooling in years.

Install

pip install juniper9-crypt

Or with uv:

uv add juniper9-crypt

Command-line usage

# Decrypt a $9$ value
juniper9-crypt --decrypt '$9$FNkC3/t1IcevLuOWx'

# Encrypt a plaintext
juniper9-crypt --encrypt 'mysecret'

# Check a $9$ value against a plaintext or another $9$ value
juniper9-crypt --check '$9$FNkC3/t1IcevLuOWx' 'hello'
juniper9-crypt --check '$9$FNkC3/t1IcevLuOWx' '$9$o1aGiPfz/Cuk.tO'

Always quote $9$ strings with single quotes - the shell expands $9 as a positional parameter otherwise.

Exit codes

Example output

$ juniper9-crypt --decrypt '$9$FNkC3/t1IcevLuOWx'
hello

$ juniper9-crypt --encrypt 'hello'
$9$o1aGiPfz/Cuk.tO

$ juniper9-crypt --check '$9$FNkC3/t1IcevLuOWx' 'hello'
Value 1   : 'hello'
Value 2   : 'hello'
Match     : YES

--encrypt output varies on every run: the algorithm inserts random filler characters, so the same plaintext produces a different ciphertext each time. They all decrypt back to the same plaintext.

Python API

from juniper9_crypt import decrypt, encrypt, check

# Decrypt
plain = decrypt("$9$FNkC3/t1IcevLuOWx")
# 'hello'

# Encrypt (non-deterministic)
ciphertext = encrypt("hello")
# '$9$o1aGiPfz/Cuk.tO'  (or any other valid $9$ form)

# Compare a $9$ value against a plaintext
plain_a, plain_b, match = check("$9$FNkC3/t1IcevLuOWx", "hello")
assert match is True

# Compare two $9$ values
plain_a, plain_b, match = check(
    "$9$FNkC3/t1IcevLuOWx",
    encrypt("hello"),
)
assert match is True

Error handling

decrypt() raises ValueError for malformed inputs (missing $9$ prefix, characters outside the alphabet, truncated ciphertext). encrypt() raises ValueError for plaintext containing characters outside Latin-1, which the cipher cannot represent:

from juniper9_crypt import decrypt

try:
    decrypt("not-a-juniper-string")
except ValueError as e:
    print(f"bad input: {e}")

Tests

git clone https://github.com/antoinekh/juniper9-crypt
cd juniper9-crypt
uv run pytest -v

Credits

I'm not nearly smart enough to have reverse-engineered Juniper's $9$ cipher on my own. This package is a small refactor on top of work done by people who actually figured it out, with packaging, tests, and the Python API cleanup done by me with help from Claude.

• Crypt::Juniper - original Perl module by Kevin Brintnall (the real reverse-engineering work)
• junosdecode - Python 2 port by Matt Hite (where the Python implementation comes from)
• This package - Python 3 port, type hints, encrypt() fix, check() API, CLI, tests, and PyPI packaging

Algorithm

$9$ is a position-based substitution cipher with three moving parts: a fixed 65-character alphabet split into families, a fixed weight table per output position, and a chain of "gaps" between successive alphabet positions.

Building blocks

1. The alphabet. 65 characters split into four ordered families:

FAMILY = [
    "QzF3n6/9CAtpu0O",          # family 0 (15 chars)
    "B1IREhcSyrleKvMW8LXx",     # family 1 (20 chars)
    "7N-dVbwsY2g4oaJZGUDj",     # family 2 (20 chars)
    "iHkq.mPf5T",               # family 3 (10 chars)
]
ALPHA = "".join(FAMILY)         # 65 chars total

Each character has a fixed index 0-64 in ALPHA. The family a character belongs to controls how much "filler" is inserted after it (see step 4).

2. The weight table. A cycle of 7 weight vectors, one per output byte position:

ENCODING = [
    [1, 4, 32],         # byte 0
    [1, 16, 32],        # byte 1
    [1, 8, 32],         # byte 2
    [1, 64],            # byte 3
    [1, 32],            # byte 4
    [1, 4, 16, 128],    # byte 5
    [1, 32, 64],        # byte 6
]

The weights for byte i are ENCODING[i % 7]. Their length (2-4) tells you how many ciphertext characters encode that byte. The weights are bases in a mixed-radix number system: an output byte b is decomposed as b = g0*w0 + g1*w1 + ... where each gk is a small "gap" value.

3. The gap. The cipher never stores absolute positions - only gaps between consecutive characters:

gap(c1, c2) = (NUM[c2] - NUM[c1]) % 65 - 1

So a gap of 0 means "the next character in ALPHA", a gap of 1 means "skip one", etc. Gaps are taken modulo 65, so the alphabet wraps.

Encryption

To encrypt a plaintext like "hi":

1. Pick a random start character from ALPHA. Call it s. Output: $9$s.
2. Insert filler. Look up EXTRA[s] (a value 0-3 depending on which family s belongs to). Append that many random characters from ALPHA. This filler is decorative - it's discarded on decrypt. It exists purely to randomize the visual appearance of the output.
3. For each plaintext byte (in order):
   • Look up the weights w = ENCODING[i % 7] for this position.
   • Decompose the byte's value into gaps using the weights:
     • gap_last = byte // w_last, remainder = byte % w_last
     • Repeat down through the weights.
   • For each gap, advance from the previous output character by gap + 1 positions in ALPHA (mod 65) to find the next output character.
   • Append those characters to the output.

Because step 1 is random, the same plaintext produces a different ciphertext every call. But because the math is fully reversible, all those ciphertexts decrypt back to the same plaintext.

Decryption

1. Strip $9$ and read the first character s.
2. Skip EXTRA[s] filler characters.
3. Walk the rest in chunks, sized by ENCODING[i % 7] for each output byte i:
   • For each character in the chunk, compute the gap from the previous character.
   • Multiply each gap by its corresponding weight and sum: byte = g0*w0 + g1*w1 + ...
   • Take byte % 256 and emit it as the plaintext character.

A worked micro-example

Encrypting the single byte 'h' (ASCII 104) at byte position 0, starting from previous character Q (index 0 in ALPHA):

• Weights: [1, 4, 32] - the byte is decomposed as g2*32 + g1*4 + g0*1.
• Decompose: 104 = 3*32 + 2*4 + 0*1 - so gaps are [g0=0, g1=2, g2=3].
• Walk forward in ALPHA, advancing gap + 1 positions each step:
  • From Q (index 0), advance 0+1 = 1 - index 1 - z
  • From z (index 1), advance 2+1 = 3 - index 4 - n
  • From n (index 4), advance 3+1 = 4 - index 8 - C
• The three output characters for the byte 'h' are znC.

On decryption, the chain Q - z - n - C is read back: gaps 0, 2, 3, recombined as 0*1 + 2*4 + 3*32 = 104 = 'h'.

Why it's weak

There is no key. The alphabet, families, and weight tables are constants baked into every Juniper device and every implementation (including this one). Anyone with the ciphertext can recover the plaintext using only public information. $9$ is obfuscation, not encryption - it stops shoulder-surfing in a config dump, nothing more.

License

MIT