Control plane, not a runtime proxy

• Status: accepted (records a decision the code already follows)
• Date: 2026-10-04

Context

Ahena holds credentials for the providers an application uses and can call their APIs. It would be technically easy to also route the application's own traffic (completions, emails, charges, uploads) through Ahena: one key, central logging, "switch providers without code changes".

Decision

Ahena manages, configures and verifies provider integrations outside the application's request path. Applications call providers directly, using code Ahena generates with official SDKs or code the developer writes. Ahena's API serves Ahena's own control surface only.

Consequences

• No added latency, and an Ahena outage does not take customer applications down.
• No runtime dependency on Ahena: generated code imports nothing from Ahena and calls nothing at Ahena. Developers can keep or edit it without Ahena.
• Ahena cannot observe runtime traffic; health comes from provider state (Doctor, drift), not request logs. Uptime monitoring is out of scope.
• Provider switching needs code changes (generated adapters keep them small) instead of a proxy flip.
• Management features (plan, apply, Doctor, the CI deploy check) do depend on Ahena being reachable; the CI check fails closed when it isn't.