Skip to main content
AI governance register

GovernSutra AI

Local AI-system register. Not a compliance certificate.

Offline register of AI systems with ten rules, evidence links, and a hash-chained audit log. npm [email protected] published 2026-11-19. The command alias is governsutra. The README says there is no API key, no cloud, and no telemetry. It gives no legal advice, assigns no regulatory category, and does not certify ISO/IEC 42001 or NIST AI RMF. Unknown is not a pass. Every assessment ends HUMAN_REVIEW_REQUIRED. On 2026-11-20 the npm downloads API returned no row, so both counts are 0. GitHub showed 0 stars and 0 open issues. The handbook home said 80 passed, 0 skipped, 0 failed. The npm README said 78 tests. This page did not re-run either. The name GovernSutra has not been trademark-searched.

  • Offline
  • 10 rules
  • Node 22+
GitHub version
0.2.1
Licence
Apache-2.0
Language
JavaScript
Weekly downloads
0
Overall downloads
0
GitHub stars
0

Use it when

  • 1

    You want a local register of AI systems, owners, models, tools, MCP servers, and data sources.

  • 2

    You want ten named rules that can return unknown instead of a pass.

  • 3

    You want a sealed Markdown, HTML, and JSON report without an API key.

What it does, and what it does not

Read this before the guide. A limit here is a limit of the tool.

Does

  • register and list keep owner, risk owner, intended use, and prohibited use in .governsutra/registry.json.
  • assess runs GOV-001 through GOV-010. The README says each rule is pass, fail, unknown, not-applicable, or exception-approved, with a reason.
  • The risk indicator has six factors. The README says a missing input scores mid-scale, not low.
  • scan and import-manifests read allowlisted files: package.json, lockfiles, requirements.txt, pyproject.toml, MCP configs, CI workflows, and governance/ai-systems/*.json. The README says the cap is 2 MiB and 200 manifests, and that symlinks and paths outside the root are skipped.
  • diff reports model, tool, data, prompt, owner, and intended-use changes. evidence add and evidence verify store a SHA-256, an expiry, and a named verifier.
  • audit verify checks the hash chain. The README says the chain detects edits and is not WORM storage.
  • serve binds 127.0.0.1, requires a bearer token of 24 or more characters, and rejects non-GET methods and any Origin header. 0 runtime dependencies. Node.js 22 or newer.

Does not

  • It does not give legal advice. The README says it assigns no regulatory category, including under the EU AI Act, and certifies nothing.
  • NIST AI RMF, ISO/IEC 42001, OECD AI Principles, and EU AI Act packs are informational. The README says ISO/IEC 42001 clause text is not reproduced.
  • A missing or stale control stays unknown. The README says file presence is not verification, and a control marked verified without hashed evidence and a named verifier fails GOV-004.
  • An exception needs a named human approver, a rationale, a scope, and a future expiry. The README says an agent approver is rejected.
  • install-ide copies instruction files for nine IDE hosts and does not overwrite unless --force is set. The README says native activation in each host is not verified.
  • The README lists dashboard, database storage, SSO and RBAC, signed attestations, an MCP server, and YAML manifests as planned and not built. The handbook home said 80 passed. The npm README said 78 tests. This page did not re-run them. A sealed report is not a certification. The name GovernSutra has not been trademark-searched.

Claims above are from the project pages and are attributed as stated there. They are not independently verified here.

How a run moves

From the GovernSutra AI README, npm 0.2.1, published 2026-11-19. It reads files you point at. It does not decide that a law is met.

  1. 1. Register

    Named owner

    Intended use, prohibited use, models, tools, MCP servers, and data sources. A person owns the record.

  2. 2. Rules

    GOV-001 to 010

    Pass, fail, unknown, not-applicable, or exception-approved. Unknown is not a pass.

  3. 3. Evidence

    Hash and expiry

    SHA-256, a named verifier, and a date. The README says an expired item becomes unknown. The seal is not a signature.

  4. 4. Gate

    One ending

    The README says every assessment ends HUMAN_REVIEW_REQUIRED. That is not an approval to ship.

How to read the result

Result It means It does not mean
Unknown The README says missing or stale evidence is reported as unknown. Not a pass and not a compliance finding.
HUMAN_REVIEW_REQUIRED The static rules finished. A named person still decides. Not a certificate and not a regulatory category.
Exception A named human approver, a rationale, a scope, and a future expiry. Not an agent approval. The README says an expired exception reopens the finding.

Wording is from the project page or README and is attributed as stated there. It is not independently verified here.

Commands and notes

  • npx [email protected] doctor local check. No network.
  • npx [email protected] audit --out ./governsutra-audit self-check bundle: JSON, Markdown, and HTML.
  • governsutra init and governsutra register --input my-system.json start a register.
  • governsutra assess --id MY-SYSTEM-1 --format md ten rules and the risk indicator.
  • governsutra report, diff, evidence, scan, install-ide, and serve listed in the README command table. serve is loopback only.

Questions

Does GovernSutra prove we comply with the EU AI Act or ISO 42001?

No. The README says it gives no legal advice, assigns no regulatory category, and certifies neither ISO/IEC 42001 nor NIST AI RMF. Mapping packs are informational. This page is not legal advice either.

Where is the handbook, and did this page re-run the tests?

https://shahvatsalofficial.github.io/governsutra-ai/ returned HTTP 200 on 2026-11-20. Fourteen handbook pages share one meta description, so this page shows that sentence on the documentation home card only. The site also publishes samples/audit/audit-report.html, which has no meta description. The handbook home said 80 passed, 0 skipped, and 0 failed. The npm README said 78 tests. This page did not re-run either, and it did not re-run the sample audit.

WhatsApp