Unsloth Studio just shipped a meaningful pre-revamp update that changes how you can wire the tool into your existing stack. The headline addition: you can now connect Unsloth to any cloud API provider, including OpenAI, Anthropic, and OpenRouter, directly from the interface. That matters if you have been running local models but want to compare against hosted ones without leaving your workflow.
The full changelog is on GitHub.
API calling goes deeper than a simple key field. Web search is built in for OpenAI, Anthropic, OpenRouter, and Kimi. Code execution is available for OpenAI and Anthropic, and Anthropic containers persist across turns so state is not thrown away between messages. Prompt caching is enabled automatically for OpenAI and Anthropic models, with the team citing 50 to 90% cost savings as a result. Image generation and editing are also included.
For local inference, the update formalizes connections to vLLM, Ollama, and llama-server (llama.cpp). API keys are now optional for those local providers. Models auto-load when you add a cloud provider, which removes a manual step that slowed down provider-switching.
Speculative decoding also got an upgrade. Auto MTP speculative decoding for MTP GGUFs now selects the best settings for your specific hardware automatically.
Non-English language support was a gap that is now closed. Japanese, Chinese, Indian languages, and others work properly for both sending prompts and receiving responses. The update includes IME composer hardening and RTL direction handling (dir="auto"), which signals this was treated as a real input correctness fix rather than a surface-level patch.
Offline use is better supported too. Cached GGUF discovery and offline DNS auto-detection now cover both inference and training workflows.
On the security side, authentication rate-limiting has been added, and it is proxy-aware so reverse proxies cannot bypass it.
For Mac users hitting issues with unsloth studio update, the team recommends using the install script directly: curl -fsSL https://unsloth.ai/install.sh | sh.
A larger revamp is coming this week or next, with new major features and significant design changes. That makes now a good time to get familiar with the current API connection setup before the interface shifts.
What to do today: if you are using Unsloth Studio for local inference and also calling cloud APIs in a separate tool, connect your API keys inside Studio now. Enable prompt caching for OpenAI or Anthropic if cost is a concern, and test web search and code execution as built-in tools rather than managing those integrations yourself.