voidstone

Data pipelines that run until they’re done.

Pull fields out of PDFs and emails, classify tickets, dedupe a CRM, label a backlog. Point the queue at your lanes and let it run until it’s done.

See plansSize your own

Pre-launch. Accounts are open; the API and checkout open soon.

How it runs.

  1. 01

    Point the queue at your lanes.

    Workers take a record, send it to a model and write the result back.

  2. 02

    Extract, classify, clean.

    Fields from PDFs and emails, labels on tickets, duplicates in your CRM.

  3. 03

    Let it run.

    A million rows take longer on fewer lanes, not more money.

Why lanes fit it.

  • Backfills are included.

    Reprocessing everything after you change a prompt is a rerun, not a new invoice.

  • Throughput is the dial.

    More lanes means more workers at once.

  • Quick models do most of it.

    Extraction and labelling run well on the one-lane models.

Works with what you use.

Every integration

LiteLLM

python
import osimport litellm response = litellm.completion(    model="openai/kimi-k3",    api_base="https://api.voidstone.net/v1",    api_key=os.environ["VOIDSTONE_API_KEY"],    messages=[{"role": "user", "content": "Write a test for this function."}],)print(response.choices[0].message.content)

Other things people run.

All use cases