Skip to main content
judgekeeper logo

judgekeeper ™

Check your LLM-as-a-judge in 5 steps.
Find. Mark. Measure. Fix. Re-check.

PyPI version Tests Apache-2.0 license

Website · Guide · Reference · Changelog

An LLM-as-a-judge is an AI model that grades another model's answers. judgekeeper shows how often your LLM-as-a-judge agrees with you, and where it goes wrong. No LLM-as-a-judge? judgekeeper can also check the answers of your chatbot, agent or any AI that answers questions (see below).

judgekeeper finds the results, opens the marking page, and shows how often the LLM-as-a-judge agrees with you.

Install

In your project folder, in its virtual environment:

pip install judgekeeper
judgekeeper

Or, with no virtual environment: uvx judgekeeper. Python 3.11 to 3.14.

Reads promptfoo, DeepEval, Inspect AI, MLflow, or your own CSV, TSV or JSONL file with input, output and verdict columns, or a file with only input and output to check your AI's answers. A verdict can be pass/fail, true/false, yes/no, correct/incorrect, right/wrong or 1/0, in any case; --label-map reads other words. Inspect AI .eval logs and MLflow need an extra: pip install "judgekeeper[inspect]" or pip install "judgekeeper[mlflow]" (the uv forms are in the reference).

Using a coding agent? Give it this page and tell it: "Follow this to check my LLM-as-a-judge." It runs the steps and gives you the marking page link; you mark, then the coding agent explains the result.

The 5 steps

  1. Find: judgekeeper finds your eval results.
  2. Mark: a page opens in your browser. Mark answers Pass or Fail; the LLM-as-a-judge's grade stays hidden. "See the numbers so far" appears with your first numbers, and "See your result" once there is a result.
    • No browser here, or a coding agent runs it? judgekeeper --no-browser prints the link to open instead.
    • Already have Pass/Fail labels? Run judgekeeper --labels your-labels.csv, or judgekeeper --labels if they are in your results. Then mark more if needed.
  3. Measure: run judgekeeper again to see how often it agrees with you.
  4. Fix: judgekeeper fix helps you write a better rule and tests it on marks kept aside. It makes no AI call: you grade the marked answers again with your own eval tool, which may use your API key.
  5. Re-check: after you change the model or the rule, grade your marked answers again (judgekeeper fix writes them to .judgekeeper/fix/marked-answers.csv). judgekeeper check <new results file> shows if it still agrees with you.
    • Changed your AI, not the LLM-as-a-judge? judgekeeper results.json --before main-results.json compares the grades in two results question by question and exits 1 when the change made it worse, or 3 when nothing could be compared (how).

In your project, judgekeeper reads your results files and writes only in .judgekeeper/, plus judgekeeper.toml after you say yes. It makes no AI call and needs no API key. See SECURITY.md.

No LLM-as-a-judge? Check your AI's answers

For a chatbot, agent or any AI that answers questions. Save its answers in a CSV, TSV or JSONL file with input and output columns and no grades. No answers yet? judgekeeper prints a prompt for your coding agent (Claude Code, Cursor or Codex) to write about 40 test questions and save your AI's answers. Your AI runs once for each question, so this may use your API credits.

  1. Find: run judgekeeper in your project folder. It finds your answers.
  2. Mark: mark each answer Pass or Fail. On the result page, add a reason to the failed ones.
  3. Measure: from 10 marks, see how many passed, with a range.
  4. Fix: judgekeeper saves the failed answers, apart from the half kept aside, in .judgekeeper/fix/failed-answers.md, with a prompt for your coding agent to improve your AI's prompt.
  5. Re-check: run your AI again and mark the new answers. On the questions kept aside, judgekeeper counts how many were fixed and how many broke, and says whether the change is more than a coincidence.

An example run.

The numbers

  • TPR: of the answers you passed, how many it passed too. The result page calls it "When you say Pass, your LLM-as-a-judge also says Pass".
  • TNR: of the answers you failed, how many it failed too. The result page calls it "When you say Fail, your LLM-as-a-judge also says Fail".
  • Kappa: how much you agree beyond what matching at random would give.

TPR and TNR each come with a range, such as "about 90% (between 85% and 93%)" (when you mark every answer, the numbers have no range). The result says "good enough to use" only when a wider range (99%) starts at 80% or more for both. That way you can look after every mark. The numbers are about the answers in your results, worked out from the ones you marked. They are not a promise about new answers.

How the numbers are worked out: reference.

Checked on public data

These runs use public datasets that people already labelled. One adds made-up empty answers that should fail, and one is a simulation. Where the true number is known, they test whether judgekeeper's ranges hold it. They also show what judgekeeper says about real LLM-as-a-judge setups. Each line links its report.

  • JailbreakBench: the ranges held the true values in 96 of 100 random tests with 100 marks each (TPR, TNR and kappa at once; another 100 tests give a slightly different count).
  • RAGTruth: news summaries. Three LLM-as-a-judge setups, 3 runs each, and 3 more runs at a lower effort setting: needs work in all 12. Changing the model did not close the gap.
  • RAGTruth question answers and data-to-text: "needs work" in all 4 runs.
  • JailbreakBench graded by two other setups: "needs work" for both; each passed more than half of the expert-failed answers it graded (setup B could not grade 61 of 300).
  • Empty answers: one model failed 0 of 20 empty summaries that should fail.
  • Simulation: the ranges held the true value 96.7 times in 100 on average.

More

A guide on the website walks through the 5 steps. judgekeeper --help lists every command, and the reference explains each one. Made by one person; issues welcome, pull requests are not taken.

License

judgekeeper and the judgekeeper logo are trademarks of Sathvik Thota. The Apache-2.0 license covers the code, not the name or logo.

Metadata

Release files for judgekeeper 1.3.0

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for judgekeeper 1.3.0
File Size Uploaded Provenance Actions
judgekeeper-1.3.0.tar.gz 635.4 kB With provenance Details for judgekeeper-1.3.0.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for judgekeeper 1.3.0
File Interpreter ABI Platform Actions
judgekeeper-1.3.0-py3-none-any.whl Python 3 none any Details for judgekeeper-1.3.0-py3-none-any.whl

Total release size: 1.3 MB

Release files / judgekeeper-1.3.0.tar.gz

Download URL judgekeeper-1.3.0.tar.gz
Size 635.4 kB
Tags Source
SHA-256 checksum
How to use checksums
1375d884b107fee18088538c7f2de34b2ab45fc3247a8bbc9498e8cc71143435
BLAKE2b-256 checksum
How to use checksums
8d8d1062e1f751de0b87a546f3a923699837760e79875a00b5ed822886d59931
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 22, 2026.

Transparency log

Release files / judgekeeper-1.3.0-py3-none-any.whl

Download URL judgekeeper-1.3.0-py3-none-any.whl
Size 657.1 kB
Tags Python 3
SHA-256 checksum
How to use checksums
8ab0241ce2d935a166fb0ce84f73f60fd020d72ac6d2aa8eeaa5753143772899
BLAKE2b-256 checksum
How to use checksums
cabb95383dc5b525965ccbd3440bb9001ea930774d0f57f65d48af2e61419413
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 22, 2026.

Transparency log

Security for judgekeeper 1.3.0

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

Full provenance

Both files in this release have been uploaded with provenance metadata.

Select from 2 files

Attestation for judgekeeper-1.3.0.tar.gz

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 22, 2026.

Transparency log

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Release history Release notifications | RSS feed

This release

1.3.0 This release

2 release files Full provenance

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page