🎉 What's New
🔧 Claude code Custom Models & API Gateways (DeepSeek, OpenAI, Kimi, GLM & More)
Settings > Claude now directly edits the env block of ~/.claude/settings.json. With one click, set up your base URLs, auth tokens, and per-model overrides — so you can easily run DeepSeek, OpenAI, or any custom API gateway inside Claude Code.
🎚️ Effort Levels for Claude Code
Claude Code now runs at the effort you choose — low, medium, high, xhigh or max. The model selector became a popover with a dedicated effort slider, shared between the codex and Claude Code selectors, and the fable model joined the list.
🔐 New Permission Modes
Three modes to match how much you want to be asked: auto, dontAsk and manual, each with matching hook behavior.
🤖 Automation Runner
Automations run on a configurable worktree. Runs orphaned by a crash are marked failed on startup, and quitting asks for confirmation while a workspace still has active work.
🎨 Theme Backgrounds
Set your own background image, or turn on the starfield effect.
🪟 Windows Portable Build
A portable archive is now published with every release, and the Scoop install path is documented in the README.
⚡ Model Selector Improvements
- Reasoning effort moved into its own selector.
- Model lists are fetched once and shared across selectors.
- Providers you don't use can be hidden.
- Provider env status is read in a single batch instead of once per key.
🐛 Fixes & Polish
- Sessions can no longer be created or resumed without a working directory.
- Dropped the
codex-protocoldependency in favor of local types, and upgraded claude-agent-sdk-rs to v0.8.0.