vxdf

VXDF Python Library

VXDF (Vector eXchange Data Format) is an AI-native container for text, metadata and vector embeddings—portable, indexable and compressed. If you do RAG, semantic search or compliance audits, VXDF gives you one file, one command.

Quick-start

pip install vxdf[zstd]        # installs optional Zstandard support

python - << 'PY'
from vxdf import VXDFWriter, VXDFReader

# create a small file
data = [
    {"id": "1", "text": "hello", "vector": [0.1, 0.2]},
    {"id": "2", "text": "world", "vector": [0.3, 0.4]},
]
with VXDFWriter("demo.vxdf", embedding_dim=2, compression="zstd") as w:
    for chunk in data:
        w.add_chunk(chunk)

# read it back
a = VXDFReader("demo.vxdf")
print(a.get_chunk("2"))
PY

Command-line

vxdf pack data.jsonl data.vxdf --compression zstd   # create
vxdf info data.vxdf                                  # header & stats
vxdf list data.vxdf | head                           # ids
vxdf get  data.vxdf some-id > doc.json               # extract

# Pipe stdin to stdout (auto-detects model, disables banner/progress)
cat report.txt | vxdf convert - - > report.vxdf

Colab / Notebook

LangChain integration (preview)

from langchain_community.vectorstores import VXDF
vs = VXDF.from_vxdf("demo.vxdf")

See examples/langchain_integration.py for a minimal adapter.

Authentication

VXDF commands that interact with cloud services need credentials.

OpenAI embeddings

The client looks for an API key in this order (first match wins):

• --openai-key CLI flag (e.g. vxdf convert my.pdf out.vxdf --model openai --openai-key sk-...)
• OPENAI_API_KEY environment variable.
• ~/.vxdf/config.toml under the [openai] table:

[openai]
api_key = "sk-..."

AWS (S3 URLs)

Uses the standard AWS credential chain provided by boto3 – environment variables (AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY), the AWS CLI config, or an attached IAM role. Run aws configure if unsure.

GCP (gs:// URLs)

Relies on Application Default Credentials. Run gcloud auth application-default login or set the GOOGLE_APPLICATION_CREDENTIALS environment variable pointing at a JSON key file.

If credentials are missing VXDF exits early with a clear message and a hint on how to configure them.

Shell completion

Install extra dependencies and activate once:

pip install vxdf[completion]
activate-global-python-argcomplete --user  # bash/zsh/fish supported

Re-open your terminal and enjoy TAB-completion for vxdf sub-commands and options.

VXDF is BSD-3-licensed. Contributions welcome!