SimpleAudit

license

MIT(OSI

source

public

core-gated

ungated(local execution is the default mode, all scenario packs ship in the package)

Permissive OSI (MIT) license with full public source. The README describes a local-first framework that ships its scenario packs and needs no account, and no paid tier, hosted edition or license key appears anywhere in it.

• https://raw.githubusercontent.com/kelkalot/simpleaudit/main/LICENSE recorded 2026-06-22

  LICENSE body 'MIT License', copyright Simula Research Laboratory

• https://www.digitalpublicgoods.net/r/simpleaudit recorded 2026-06-22

  DPG profile confirming multilingual safety red-teaming framework

• https://raw.githubusercontent.com/kelkalot/simpleaudit/main/README.md recorded 2026-08-11

  README states 'Local execution is the default deployment mode and the original design constraint: prompts, transcripts, and policies stay inside the deployment environment' and 'MIT License - see LICENSE for details'. The scenario packs and judge configurations ship in the package; the only external dependency is the user's own model API key.

• https://raw.githubusercontent.com/kelkalot/simpleaudit/main/LICENSE recorded 2026-08-13

  LICENSE body is 'MIT License', copyright Simula Research Laboratory.

• https://raw.githubusercontent.com/kelkalot/simpleaudit/main/README.md recorded 2026-08-13

  'SimpleAudit is a simple, extensible, local-first framework for multilingual auditing and red-teaming of AI systems via adversarial probing. It supports open models running locally (no APIs required)', and it closes 'MIT License - see LICENSE for details'. The scenario packs, rubric, auditor and judge configuration ship in the package; the only external dependency is the reader's own model API key if they choose to use one.

kelkalot/simpleaudit carries 20 GitHub stars and 9 forks, and describes itself as early-stage. That bands at fewer than 1K stars, level 1; stars cap a level at 3 in any case, because a star is not a use. No download, install or customer figure is published for this product, so stars are the only honest signal available and the level is directional.

• https://api.github.com/repos/kelkalot/simpleaudit recorded 2026-08-12

  stargazers_count = 20 for kelkalot/simpleaudit

A focused but functional red-teaming toolkit with a distinctive broken-premise mode. The README carries a validation section reporting AUROC of 0.89-1.00 for separating safe from unsafe responses, and a direct comparison against Petri, but this is still a single-purpose adversarial-probing tool rather than a general harness, which is what holds the score where it is.

• https://github.com/kelkalot/simpleaudit recorded 2026-06-22

  README: adversarial probing, local-first, scenario packs, broken-premise runner

• https://raw.githubusercontent.com/kelkalot/simpleaudit/main/README.md recorded 2026-08-13

  multilingual adversarial probing with local open models or optional API-hosted ones; a fixed default scenario pack, rubric, auditor, judge and sampling configuration so reruns are comparable; reported responsiveness of AUROC 0.89-1.00 across judge-auditor cells; results reported as a bundle rather than a leaderboard rank.