June was about making ModelRiver easier to trust in production and easier to explain to teams evaluating AI infrastructure. We added more provider choice, improved safety controls, and tightened the product experience across the console, auth, docs, and marketing pages.
New providers: Kimi and MiniMax
ModelRiver now supports Kimi and MiniMax as provider options. Teams can add these alongside existing providers and use them in workflows where model quality, latency, pricing, or regional strengths make them a better fit.
This keeps the workflow model flexible: connect the provider, pick the model, and keep the application integration stable.
Centralized input guardrails
We added centralized input guardrails with local and remote moderation support. Guardrails now cover more entry points and classify risky prompts before they move deeper into a workflow.
The first version focused on practical safety controls:
- Harmful prompt detection across supported request paths
- Risk tiering for input moderation
- Better detection for paraphrased harmful instructions
- Stronger handling for weapon and crime-instruction patterns
- Clearer provider error messaging when a request is blocked or fails
Model switching guidance
We published model switching guidance for teams that want to change AI models without rewriting product code. The core idea is simple: keep your app calling a stable ModelRiver workflow, then switch providers or models behind that workflow as needs change.
Read the full article: Switch AI models without rewriting your product.
Product polish
We unified the visual language across the console, auth pages, blog, docs, and marketing pages. The new stone theme, dark mode polish, system appearance mode, narrower modals, cleaner menu dividers, and updated switcher badges make the product feel more consistent from first visit to daily console usage.
We also improved the sidebar project switcher and request log duration formatting, making it easier to move between projects and scan request timelines.