RELEASE NOTES
Changelog
-
Spend limits, and apps you can manage
An organization can cap what it spends per month and what each of its apps spends per day, and the console can create, show and suspend apps.
Read the release notes → -
Tool schemas with a top-level anyOf
onde-large, onde-balanced and onde-fast no longer reject a request because one tool's schema starts with oneOf, anyOf or allOf.
Read the release notes → -
Images in chat messages
A message can now carry images, on the tiers whose models can read them.
Read the release notes → -
A new tier for coding and agentic work
onde-prism joins the catalogue with two providers behind it from the first day.
Read the release notes → -
Each model's maker first
Tiers now try the API of the company that built the model before any third-party host, onde-zenith gains a second provider, and requests that force a tool call skip providers that cannot honour them.
Read the release notes → -
Cheaper multi-turn chats and more places to fail over
Onde Cloud now keeps each conversation on a warm prompt cache, reports cached tokens, adds two new tiers, and gives more tiers a second provider to fall back to.
Read the release notes → -
Knowing when a provider is out
Onde Cloud now tells an empty or broken provider account apart from a bad request, stops sending traffic to it, and tells the operator.
Read the release notes → -
Provider failover and a clearer catalogue
Onde Cloud now has an ordered provider engine, Baseten support, bounded upstream calls, and release notes for the product taxonomy behind the router.
Read the release notes → -
First public release
Onde Cloud launches as an OpenAI-compatible inference API with authenticated model routing and usage accounting.
Read the release notes →