The data behind the site, as files you can build on.

The envelope

Every file wraps its data in the same top-level fields:

• notice: the disclaimer above, with the Body of Knowledge version.
• version: the Body of Knowledge version the file was built from (0.5.0).
• license and licenseUrl: CC BY 4.0 and its deed.
• schemaVersion and schema: the shape version and the JSON Schema the file validates against.
• self and source: the file's own URL and the page that renders the same data.
• citation: the title, authors and DOIs to cite (see below).

Schemas are JSON Schema draft 2020-12, one per file, at https://aigovernanceengineer.com/api/v1/schemas/<name>.json: obligations , frameworks , crosswalk , glossary , patterns , maturity , path , chapters , jurisdictions , harms , cases , contracts , roles , threats , controls , obligation , control , index . Every record is a closed object: a field is always present, and absent values are null or an empty list, never missing.

Obligation ids

Each obligation carries an id of the form AIGE-OBL-<INSTRUMENT>-<CLAUSE>, for example AIGE-OBL-EUAIA-ART9. Ids use upper-case letters, digits and hyphens only (the pattern is ^AIGE-OBL-[A-Z0-9]+(?:-[A-Z0-9]+)+$); URLs use the same id in lower case.

• An id is assigned once. New wording, a moved date, a better artefact or a changed layer keeps the id.
• A removed row keeps its id reserved; it is never reused for another obligation.
• A split row keeps the id on the part that carries the original meaning; the new part gets a new id.

The instrument codes in use:

Versioning and stability

• The path is the contract. Everything under /api/v1/ keeps its URL and its meaning for as long as v1 exists. A breaking change ships under a new path, and the old one stays up alongside it.
• Fields are added, not taken away. Within a schemaVersion, a file may gain fields; no field is removed, renamed or retyped. A breaking change to one file bumps its schemaVersion and is recorded in the changelog.
• The data moves with the book. Files are rebuilt on every release; the version field says which edition of the Body of Knowledge they come from, and each obligation carries the date it was last reviewed.
• The older downloads stay. /resources/obligations.json and its CSV keep their URLs. They moved to schemaVersion 2 in v0.5.0: appliesFrom became an ISO date (or null) and the former free text moved to appliesNote. The CSV keeps its first seven columns in place and appends the new ones.

Access

The files are static. They are served with Access-Control-Allow-Origin: *, so a browser application on any origin can read them, and with Cache-Control: public, max-age=3600. There is no key and no authentication. Please cache what you fetch; the data changes with releases, not by the minute.

# The catalogue
curl -s https://aigovernanceengineer.com/api/v1/index.json

# One obligation, by its id in lower case
curl -s https://aigovernanceengineer.com/api/v1/obligations/aige-obl-euaia-art9.json

# Every EU AI Act obligation that applies on a given date (jq)
curl -s https://aigovernanceengineer.com/api/v1/obligations.json \
  | jq '.obligations[] | select(.frameworkId == "eu-ai-act" and .appliesFrom != null
        and .appliesFrom <= "2027-12-02") | {id, appliesFrom, appliesStatus}'

Other files that predate the API and stay where they are:

• /resources/crosswalk.json and CSV
• /resources/harms.json
• /glossary.json (short definitions for hover cards)
• The governance templates and their JSON Schemas on the templates page

Remote MCP server (read-only)

The same data is served to AI assistants by a read-only Model Context Protocol server: it answers questions about the obligation register, the crosswalk, the glossary, the patterns, the templates and schemas and the chapters, and every answer names the page it was read from. It holds no data of its own, needs no account and stores nothing about the people who call it. Its endpoint is https://mcp.aigovernanceengineer.com/mcp (Streamable HTTP, no authentication), live since 2026-09-25; see the MCP server page for how to connect a client. The server's code, its tools and a guide to run it yourself are in tools/mcp-server of the repository.

claude mcp add --transport http aige https://mcp.aigovernanceengineer.com/mcp

How to cite

The data is part of the archived Body of Knowledge, not a separate deposit. Cite the release DOI 10.5281/zenodo.22956197 (this version) or the concept DOI 10.5281/zenodo.22857084 (always the latest), and name the file and the date you fetched it. For a single obligation, cite its id and its page: every obligation page carries its own citation block. Attribution under CC BY 4.0: Jorge García Aibar.