Vernia
Vernia Spendlint is live, and free

Predictable LLM spend, from pull request to invoice.

AI products get expensive quietly: a cache that never hits, a loop with no limit, a cost report that misses half the tokens. Vernia builds the tools that catch it early and show you where every dollar goes.

Get Spendlint, free See what’s coming

Products

One goal: no surprises on the bill. Start with Spendlint today.

Coming soon

Vernia Reason

A reasoning harness that reviews pull requests the way your best engineer would. It reads the whole repository, proposes deeper fixes than any linter, and gets sharper with every review.

Get early access
Coming soon

Vernia Ledger

An autonomous agent that connects to your model providers, traces and billing. It ties every dollar to a feature and a customer, flags anomalies before the invoice, and helps you plan the roadmap around real unit economics.

Get early access

Vernia Spendlint

Up and running in under a minute. Python and TypeScript, no account needed.

  1. Install

    pip install spendlint

  2. Scan your repository

    Run it from the root of any project that calls an LLM.

    spendlint .

  3. Check every pull request

    Add the GitHub Action. It flags only the lines a pull request adds, so old code never gets in the way.

    # .github/workflows/spendlint.yml
on: pull_request
jobs:
  spendlint:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v7
      - uses: vernialabs/spendlint@v0
Missing prompt cachingRepeat calls that re-pay for the same system prompt.
Volatile system promptsTimestamps and ids that break the cache on every request.
Runaway loops and retriesAgent loops and retry wrappers with no limit.
Cost reports that miss tokensAccounting that ignores cached reads and writes.
Prompt bloatPretty-printed JSON and one-at-a-time embeddings.
Free during early accessLive

Your code never leaves your machine. Usage telemetry is anonymous: counts only, never code, file names or repositories. A Teams plan with telemetry off and private rule packs is on the way.

GitHub PyPI Docs

Runs where your code lives

Spendlint scans on your laptop or your CI runner. Nothing is uploaded.

Quiet by default

On pull requests it only speaks up about the lines you changed, so reviews stay focused.

Built from real fixes

Every rule comes from cost bugs that open-source agent teams actually found and fixed.

Running agents in production?

We’d love to hear where your LLM spend surprises you. Early access to Reason and Ledger starts there.

Get in touch