Cline Desktop v0.0.40: Reliable Custom Providers, Better Cloud Switching
Cline Desktop v0.0.40 fixes longstanding provider issues, refines cloud/chat model switching, and refreshes default model lists. This improves workflow flexibility and reliability for developers using multiple AI model backends.
What changed?
Cline Desktop v0.0.40 introduces several important reliability and workflow updates: - Custom providers added via "Add Provider" now work as expected during task runs. Previously, these failed with an 'Unknown or disabled provider' error. - Chats now remember whether you last used 'Cloud' or 'Local' backends, and which specific cloud model you selected. Switching a thread from Local to Cloud also opens on your previously chosen Cloud model, instead of defaulting to Local. - Saving provider credentials is now robust even if the provider's model list fails to fetch. The model list refreshes when related keys or endpoints change. - MCP (multi-provider) settings always use the same file, regardless of how the settings path is set, eliminating previous file path inconsistencies. - Tool diffs will now respect the app's chosen font size, not a fixed size. - For tools returning output too large for the context, the agent can read beyond the cutoff, offering a preview and a paged link for full review. - The model catalog refreshes several defaults and recommended models. Vultr model IDs were renamed upstream, so previously pinned Vultr models may need to be reselected.
Why does it matter to an everyday developer?
These improvements address concrete workflow and reliability pain points for developers integrating multiple AI providers into their workflow: - Custom provider support is now reliable, allowing developers to experiment with or deploy additional LLM providers without unexpected errors. - Remembering the last used backend (Cloud or Local), and model, saves time and context-switching friction—especially valuable for workflows that swap between local inference and hosted LLM APIs. - Consistent settings and credentials handling mean fewer surprises and smoother multi-provider operations, important for iterative development and debugging. - The large-output reading feature prevents loss of information from long tool responses, improving usability in real-world development scenarios.
