February was one of the biggest months for developer experience. We improved testing, caching, request logs, OpenAI compatibility, documentation, and the public blog experience.
Test Mode for AI workflows
We focused the product story around Test Mode: the ability to build and test AI workflows without spending tokens on every development or CI run.
Test Mode lets a workflow return predefined sample data while still exercising the real ModelRiver path around authentication, routing, logging, and response formatting. It is useful for frontend development, CI pipelines, loading states, and application logic that should not depend on live model variability.
Read more:
Workflow response caching
We added workflow response caching and cache visibility in the console. This helps teams reduce repeated provider calls, improve response times, and understand when cache behavior is affecting a request.
Related improvements included cache classification in request logs, cache dashboard work, and UI changes that make cached requests easier to inspect.
OpenAI-compatible API
We added OpenAI compatibility support so teams can integrate ModelRiver with tools and applications that already speak the OpenAI-style API shape.
This reduces migration friction for teams that want ModelRiver's workflow, observability, routing, and provider controls without rewriting every call site immediately.
Request log improvements
Request logs became easier to scan and debug:
- Improved filters and tab UI
- Copy buttons for important values
- Better request detail routing
- Cached request handling
- Playground requests logged into request history
These changes make request logs more useful as a production debugging surface, not just a raw event list.
Documentation and blog improvements
We reorganized docs around observability, improved the chatbot example, added caching documentation, and launched a blog layout that matches the rest of the public site. The docs and blog work made it easier to explain ModelRiver's core ideas without relying on a live demo.