Policies
There is no standalone Policy object in Alien Giraffe. Access control is not written as a declarative YAML/JSON file that is applied ahead of time — it is expressed by creating a request (optionally from a reusable request template) that names the datasource, the scope of data, the reason, and the risk level, and then having that request approved before it can be activated.
This page exists to show how the five real components combine to produce that model. Think of it as the rulebook shape, not a file you write.
The five components, for real
Section titled “The five components, for real”- Subjects — who is asking. Every request is tied to an authenticated user, resolved through the configured
[auth]provider (local,auth0, orworkos) and a role (user,admin,super_admin). - Resources — what they’re asking for. A request’s
scopenames one or more registered datasources and, within each, specific datasets (tables, files) and columns, built from the datasource/catalog registry. - Constraints — when and for how long. Each request carries a
risk_level, avalidity_period, and an admin approval step (approve/deny) before activation; once active, it can berevoked, and it expires atvalid_until. - Channels — how the approved scope is actually reached: the Data Explorer web UI or the request-scoped query API, delivered either over the internal network or through a temporary data tunnel.
- Context — why. Every request requires a
business_justificationanddescription, and can optionally carryexternal_ticket_id/external_ticket_urlto link back to a ticketing system, plus a fullaudit_logstrail of what happened.
How a request expresses all five at once
Section titled “How a request expresses all five at once”A request (or the template it’s created from) is a single object that names all five dimensions in one payload:
{ "title": "Support request for account review", "description": "Investigate billing mismatch reported by customer", "business_justification": "Customer support needs temporary access to investigate a production issue", "risk_level": "medium", "external_ticket_id": "SUPPORT-1234", "external_ticket_url": "https://issues.example.com/SUPPORT-1234", "scope": [ { "datasource": "production-postgres", "datasets": [ { "target_location": "public.accounts", "target_name": "accounts", "columns": ["id", "email", "status"] } ] } ]}scope(built fromContractDatasourceInput/ContractDatasetInput) is the resources dimension.risk_leveland the approval it triggers, plus the request’svalidity_period/valid_until, are the constraints dimension.business_justification,description, and the optionalexternal_ticket_*fields are the context dimension.- The identity that submitted and later approved the request is the subjects dimension.
- Once activated, the request exposes a channel (Explorer, API, or a temporary tunnel) scoped to exactly that request.
There is no separate step where you author a policy that then matches future requests. Reusable shape comes from request templates (POST /v1/request-templates), which pre-fill the same fields so requesters don’t redefine scope and justification from scratch every time — see Create A Data Access Request From A Template.
Where each piece is enforced
Section titled “Where each piece is enforced”- Enclave requests (the default
type) copy the approved scope into an isolated, ephemeralenv-enzoruntime and sever the upstream connection before opening it for queries — see Data Enclaves. - Proxied requests keep a live, policy-enforcing connection to the datasource instead of copying data, rewriting and filtering every query against the request’s scope in real time — see Constraints and
connector/docs/feature-proxied-request-type.md.
Related Components
Section titled “Related Components”- Subjects — identity, roles, and the auth providers a request’s actor is authenticated through
- Resources — datasources, the catalog, and how scope is built
- Constraints — risk level, validity period, approval, and the enclave/proxied session model
- Channels — how approved access is actually delivered
- Context — justification, ticket links, and the audit trail