DarkVault Python SDK

A synchronous Python client for DarkVault. Read configuration buckets, manage individual secrets, and use optimistic concurrency without implementing JWE yourself.

Install

python -m pip install darkvault-client

Requires Python 3.11+. The package name is darkvault-client; the import is darkvault. The direct runtime dependency is jwcrypto.

Quick start

import os

from darkvault import DarkVaultClient

with DarkVaultClient(
    "https://vault.example.com",
    os.environ["DARKVAULT_TOKEN"],
    timeout=10,
) as vault:
    settings = vault.read_bucket("app_prod")
    database_url = settings["Database:Url"]

The result is a dict[str, str]. Case, Unicode, empty strings, and newlines are preserved. read_bucket_snapshot additionally returns the bucket ID and revision.

Connection and permissions

Use the server's HTTPS origin, for example https://vault.example.com, without an API path, URL credentials, query, or fragment. Tokens are issued in the DarkVault web UI and use the dv1_ format (64 characters total). Supply the token explicitly from a protected source; this SDK does not load CLI configuration.

Bucket arguments are exact names, not UUIDs. Each operation needs its matching token scopes and bucket access. Reading an entire bucket needs bucket:read, secret:read, and secret:list. Bucket creation additionally depends on the token's allowed creation names. Token issuance and revocation are administrative operations, not SDK methods.

Values are strings. Treat returned secrets as sensitive in-memory data; do not log values, Authorization headers, or request bodies. TLS verification must stay enabled. JWE uses ECDH-ES/P-256/A256GCM in addition to HTTPS; the server still sees plaintext secrets, so this is not a zero-knowledge system.

Constructor and lifetime

DarkVaultClient(server, token, *, timeout=30, ssl_context=None)

timeout is a finite number of seconds greater than 0 and at most 30. It covers discovery, retries, and the command together. OS hostname resolution is subject to the platform resolver's behavior.

Reuse a client, preferably with with; alternatively call close() when finished. Calls on one client serialize over a shared HTTPS connection. Use one client per worker for parallel I/O. This SDK is synchronous, not an asyncio client. Invalid constructor settings raise ValueError; a closed client cannot be reused.

Required scopes

Scopes do not replace bucket grants.

API reference

Names, keys, and descriptions are strings. Secret writes accept string, int/float, bool, or None; ordinary reads expose canonical strings and type metadata. Revisions are integers from 0 through 9007199254740991. List options and recursive are keyword-only.

Metadata is returned as dictionaries with camelCase keys:

• Bucket: id, name, description, revision, createdAt, updatedAt.
• Secret metadata: id, bucketId, key, revision, createdAt, updatedAt, type.
• Page: items, nextCursor.
• Token: id, name, scopes, bucketIds, allBuckets, creatableBucketNames, expiresAt.

Dates are UTC RFC 3339 strings; expiresAt may be None.

Create, update, and delete

Inside a with DarkVaultClient(...) as vault: block, with an existing bucket:

created = vault.add_secret("app_prod", "Feature:Enabled", "true")

updated = vault.update_secret(
    "app_prod",
    "Feature:Enabled",
    "false",
    expected_revision=created["revision"],
)

vault.delete_secret("app_prod", "Feature:Enabled", updated["revision"])

A stale revision returns revision_conflict. Read current state before deciding whether to retry. set_secret with revision 0 creates only when absent. Deleting a nonempty bucket needs recursive=True; fetch fresh bucket metadata after changing secrets before deleting the bucket.

Pagination

Inside the same client context:

cursor = None

while True:
    page = vault.list_secrets("app_prod", cursor=cursor, limit=100)

    for item in page["items"]:
        # Metadata only.
        print(item["key"])

    cursor = page["nextCursor"]

    if cursor is None:
        break

Limits are 1–200. Treat cursors as opaque and use them for the same listing. The SDK returns one page per call.

Errors and retries

from darkvault import DarkVaultError

try:
    settings = vault.read_bucket("app_prod")
except DarkVaultError as error:
    print(error.code, error.status, error.request_id)

Run this inside the live client context. DarkVaultError carries safe metadata, not credentials or raw payloads. Status is 0 for a local/transport failure. Server errors include unauthorized, forbidden, not_found, already_exists, revision_conflict, and bucket_not_empty.

Reads may retry twice on network failures and plaintext 429/5xx responses. Retry-After is honored within the deadline. An explicit unknown_key response permits one key refresh. Uncertain writes are not retried: on outcome_unknown, read the server state before repeating the mutation.

Private certificates and networking

import ssl

context = ssl.create_default_context()
context.load_verify_locations(cafile="/path/to/company-ca.pem")

with DarkVaultClient(
    "https://vault.example.com",
    token,
    ssl_context=context,
) as vault:
    settings = vault.read_bucket("app_prod")

Here token is the token loaded from your protected source. The CA file contains public trust certificates, not your client token. Disabling certificate or hostname verification is rejected. Redirects and plaintext success responses are rejected. The client connects directly and does not inherit proxy/credential environment settings.

Local installation

From a source checkout: python -m pip install ./clients/python. For consumer applications, use a published PyPI version and pin it through your normal dependency management.

DarkVault and server setup · API schema · Issues · MIT license

Typed configuration

Secret types are string, number, boolean, and null. String reads remain available; typed reads preserve scalar types. See the configuration contract for SDK methods, nested paths, JSON/YAML export, and string fallback rules.