Skip to content

Constraints

Constraints represent when and for how long access is granted. In Alien Giraffe this isn’t a separate policy object — it’s a set of real fields and states on the request itself.

Every request carries its own temporal and approval boundaries directly, rather than inheriting them from a matching policy — see Policies for how this fits with the other four components.

draft → pending approval → approved → active → revoked / expired
  • risk_level (low | medium | high) is set when the request (or its template) is created, and is one of the signals an approver reviews.
  • validity_period (on the template) and valid_until (on the request) bound how long the request, once activated, stays usable.
  • Approval — an admin reviews and either approves or denies the request before it can be activated:
    • POST /v1/admin/requests/{id}/approve
    • POST /v1/admin/requests/{id}/deny
  • Activation — POST /v1/requests/{id}/activate provisions the runtime for an approved request and moves it to active.
  • Revocation — POST /v1/requests/{id}/revoke tears the runtime down and ends access before valid_until, if it’s no longer needed.
  • Expiry — once valid_until passes, the request’s access surfaces (query endpoint, Explorer, any tunnel) stop working, without a separate manual step.

There is no self-service CLI or chat command that grants access directly — every request goes through this same approve-then-activate path. See Approve And Manage Requests for the admin-side flow and Create A Data Access Request From A Template for the requester-side flow.

The request’s type field ("enclave" by default, or "proxied") decides how the approved scope is actually enforced once the request is active:

On activation, env-enzo pulls the approved scope into an isolated, ephemeral DuckDB instance and severs the upstream connection — see Data Enclaves. This gives strong isolation and a one-time snapshot, at the cost of not reflecting later changes to the source data.

No data is copied. A dedicated, per-request proxy container keeps a live connection to the datasource and enforces the request’s scope on every query in real time — rewriting the outbound query and filtering the response — rather than working from a static snapshot. This is useful when the source schema may drift during the request’s lifetime, or when a full copy of the data shouldn’t exist anywhere. Column-level rules (block or deidentify, with mask/hash/redact/partial strategies) can be attached at the system, datasource, or dataset level — most specific wins. A global setting controls how a blocked column is handled if it’s referenced in a query: strict (reject the query outright) or permissive (default — strip the column from the result). Proxied requests currently support the postgres and mysql dialects.

Both request types share the same lifecycle above (approve → activate → revoke/expire) and the same channels for delivery — only what runs behind activation differs.

  • Policies — how a request’s temporal bounds combine with the other four components
  • Subjects — who can approve, deny, or revoke a request
  • Resources — the scope that a request’s constraints bound access to
  • Channels — how an active request is actually reached
  • Context — the justification an approver reviews alongside risk_level