Blog
From the team building APEXVYRA
Engineering notes, product decisions, and the occasional postmortem. Written by the people who shipped it.
Why APEXVYRA routes requests instead of picking one model
Picking one model optimizes for today's benchmark. Routing optimizes for the request in front of you.
A practical guide to prompt versioning
Prompt versions are diffable and reversible. Here's how to structure them so a bad change never ships silently.
How guardrails evaluate outputs before they ship
A guardrail isn't a filter bolted on after the fact. Here's where checks run, and why order matters.
RAG in production: lessons from embedding cache misses
A cache miss just looks like a slow request. Here's how we found ours.
Idempotency keys: why retries need them
A retried request without an idempotency key is a second, different request. Here's what that costs you.
Announcing the Go SDK
A first-party Go client, built against the same API surface as Python and Node.
Designing an audit log teams actually read
Most audit logs get written and never read. Here's what we changed.
No posts in this category yet.
Stay in the loop