Domains

A domain is a knowledge scope. It defines what curated knowledge Tessera should ingest and how, so that agents can reason against grounded, versioned facts instead of guessing. A domain's output lands in the catalog, which agents read from and cite.

What a domain contains

Each domain is a directory under app/content/domains/<domain>/ with a catalog.yaml that declares its ingest pipeline:

ingest:
  sources:
    - name: au-ai-ethics-principles
      produces: [rai-principle]        # the entity type this source yields
      fetch:
        transport: static              # static | http
        data:                          # inline records for a static source
          - principle_id: fairness
            name: Fairness
            frameworks: [au-ai-ethics, nist-ai-rmf]
            # ...
      normalize:
        mode: agent                    # an agent normalizes records
        entity_type: rai-principle
        key: "{principle_id}"
        key_field: principle_id
        guidance: >-
          Emit one rai-principle record per item, preserving every field...
        ground: true
        provenance: {ingested_by: catalog-ingest, sources: [au-ai-ethics-principles]}

  refresh:
    schedule: "rate(90 days)"          # how often to re-ingest
    sanity_drop: 0.5                   # guard against a source collapsing
  synthesis: none                      # or a synthesis stage (see Catalog)

Sources

A source describes where knowledge comes from and what it produces:

• transport: static — the records are authored inline in catalog.yaml (used when the knowledge is derived, e.g. a crosswalk that isn't on any single web page). Ingest is deterministic — no fetching.
• transport: http — the records are fetched from a URL.

Normalize

normalize.mode: agent has an agent turn raw source records into clean, keyed entities of entity_type, following guidance. ground: true grounds the normalization against the source so records aren't invented. key / key_field give each entity a stable identity.

Refresh

refresh.schedule (e.g. rate(90 days)) controls periodic re-ingest. sanity_drop guards against a bad refresh — if a source suddenly yields far fewer records, the refresh is rejected rather than silently shrinking the catalog.

Shipped domains

Tessera ships domains covering Australian financial-services and AI-governance knowledge, plus platform-support domains:

Domain	Focus
accounting-tax-au	Accounting & tax guidance (AU)
au-financial-advice	Financial advice (AU)
conduct-people-au	Conduct & people obligations (AU)
cyber-security-risk	Cyber-security risk
fcc-au	Financial crime compliance (AU)
financial-regulatory-risk	Financial regulatory risk
privacy-au	Privacy (AU)
responsible-ai	Responsible-AI principles (cross-cutting)
scrutineer	Personas used by quorum deliberation
tessera-self	Knowledge about Tessera itself
schemas	Shared entity schemas

Domains vs the catalog

• A domain is the recipe — what to ingest and how.
• The catalog is the result — the grounded, versioned entities agents actually read.

Running a domain's pipeline (make ingest) populates the catalog. See Catalog for that lifecycle, and Create a new domain to author one.