News

v0.28.2 adds Nebius as a supported LLM provider across the API and desktop app, making Kimi and Qwen models available and shifting the recommended default model to Nebius Kimi K2.5 Fast. It also hardens desktop LLM configuration handling so invalid saved provider or model selections are replaced with a working recommended choice.

Highlights

• Added Nebius as a supported LLM provider in both the API and desktop app.
• Added Nebius-hosted models: moonshotai/Kimi-K2.5-fast, moonshotai/Kimi-K2.5, and Qwen/Qwen3-Embedding-8B.
• Switched the recommended default model to moonshotai/Kimi-K2.5-fast.
• Added direct and proxied Nebius routing through the OpenAI-compatible LLM endpoints, including usage accounting and tracing.

Fixes and Improvements

• Desktop startup now detects invalid saved LLM providers or models and replaces them with the current recommended provider/model instead of leaving the app on a broken selection.
• LLM settings now expose Nebius in provider selection and support storing a Nebius API key.
• Model naming and display handling were updated so Nebius Kimi and Qwen models render cleanly in the UI.
• Anonymous daily token limits were reduced from 2,000,000 to 20,000.
• Added a public status page link to the landing site footer.

Internal

• Added Nebius environment/config support in the API, desktop provider layer, pricing tables, and test coverage.