Changelog

Unreleased

Added remote API timeout fallback with a configurable model order. The default order is Kimi-K2.6, MiniMax-M2.7, Qwen3-Coder, Qwen3.6-Plus, Qwen3-Max, and Qwen3-Thinking.
Fallback only runs before any streaming bytes are sent and only uses models returned by the active /v1/models response.
Changed the default request timeout from 120 seconds to 300 seconds.
Added a desktop Settings switch and fallback model list editor.
Added persistent desktop app state for request history, app logs, and cumulative token usage.
Added a Dashboard token usage card and model-list specification chips for context window and capability summaries.
Added model display to the desktop request stream table and model-aware request search.
Fixed Dashboard "recent model" tracking so health/model-list requests no longer override the last real chat model.
Updated architecture documentation to cover the IPC and Remote API dual-backend design.
Disabled desktop Inspector and default context menu in production builds; local development can opt in with LINGMA_DESKTOP_DEBUG=1.

Default backend changed to remote API mode for new CLI and desktop configurations.
Default model changed to kmodel (Kimi-K2.6 in Lingma remote model list).
Removed the proxy-injected fake Auto model in remote mode so the model list only shows models returned by Lingma.
Fixed Dashboard recent requests showing MiniMax-M2.7 for model discovery and health/debug requests that do not contain a model field.
Added request record model and payload size fields for the desktop app request table.
Updated Dashboard transport display to show Remote API when remote backend is active.
Updated Hermes local config to use Lingma Proxy with kmodel and remote model IDs.
Updated README / README.zh-CN for remote-first mode, Kimi recommendation, package selection, protocol support, and debug/log endpoints.

Improved remote enterprise endpoint detection from Lingma logs.
Added support for showing detected remote base URL and credential source in desktop Settings.
Added macOS DMG packaging in GitHub Actions.

Added experimental remote API backend alongside the original IPC plugin backend.
Added remote credential import from local Lingma login cache or explicit credential files.
Added OpenAI / Anthropic compatible routing over the remote backend.
Added request and log debug endpoints for troubleshooting.