Blog
Notes from building a model gateway that can explain itself.
Architecture, billing, and privacy decisions—written for developers who need the mechanism, not the pitch.
01
Latest note
A model ID should be an API contract, not a routing hint
The architecture behind stable publisher/model IDs, deliberate route activation, and provider-neutral failures.
Read the noteThe notebook
More from the system
02
Prepaid AI billing without mystery charges
03How reservations, settlement, immutable price snapshots, and credit buckets make variable-cost requests explainable.
Useful model API logs do not need your prompts
04A content-blind observability model for debugging delivery and billing without creating a second store of private conversations.
Retries should repeat the result, not the charge
Why model requests need canonical idempotency keys, request fingerprints, and deterministic recovery across billing and inference.