Skip to main content

Concepts

Short reference for terms used across these docs.

Users and roles​

  • Developer — makes API calls, creates own API keys, creates and manages Knowledge Bases
  • Admin — everything a developer can, plus manages users and access, and reviews audit records
  • API key — a token for programmatic access (ygg_…). Holds scopes, not roles

Models​

  • Model — a specific model instance available in the catalog (e.g. advania-chat-…). Listed via GET /v1/models
  • Model catalog — the full set of models available to your account

Guardrails​

  • Guardrail — the platform's runtime policy layer that classifies each request and enforces routing to an approved model
  • Sentinel — the Advania-built guardrail that enforces sensitivity-tier routing
  • Decision — one guardrail outcome for one request. Referenced by x-decision-id in the response headers
  • Policy version — the version of the guardrail policy that produced a decision

Knowledge and files​

  • Knowledge Base (KB) — a collection of indexed content used to ground chat responses
  • Document — a file uploaded to a KB
  • Citation — a reference in a chat response back to the source content that supported a claim
  • Connection — a link to an external source (S3, SharePoint) that syncs content into a KB

Audit records​

  • Audit log — an immutable record of significant events on the platform (auth events, mutations). See Observability, cost & audit for what's captured.
  • Guardrails decision trail — a separate log that captures every chat-completions routing decision, referenced by x-decision-id. See Guardrails contract.

The two are distinct: audit logs cover platform actions; the guardrails decision trail covers request-level routing.

Rate limits​

  • RPM / TPM — request-per-minute and tokens-per-minute limits. Applied per model and per user
  • x-ratelimit-* — response headers exposing current usage vs cap

See Quotas, limits & pricing.