What's new across WAVE, at the platform, workstream, and product level. Shipping in the open.
Filtered by tag dispatch · clear
Set your organization to flag, block, or redact, and WAVE Dispatch screens every request before it reaches a model. A flagged request carries the detected classes in a response header. A blocked request returns a 400.
Call dispatch with an optimize_for mode of cost, balanced, or intelligence to steer which model handles a request. Available modes are configured per catalog and per team.
Tool-heavy streaming requests no longer cut off mid-response on a slow start. A model route pointing at a removed backend now falls through to a working provider.
The large-context tier now routes to a model that scored higher in our internal evaluation, improving answer quality for that class of request.
Four top frontier models join the customer-selectable catalog. WAVE's own routing classifier is now live, automatically tiering each request.
The routing ladder now includes additional model providers, including a priority path for review-heavy tasks. A request that fails on one provider now has more places left to try.
A new backend lets WAVE dispatch calls reach a broad marketplace of frontier models, with a signed record produced for every call. Available with a customer key or WAVE-hosted access.
Each paid WAVE Dispatch response carries a payment receipt and a signed context attestation, and you can verify them at a public endpoint.
The JavaScript SDK can now verify a signed record without a network call or a pre-shared key, entirely offline. It also flags when a prompt was truncated.
Dated, versioned changes to the API and SDKs live at changelog.wave.online.