2.0.32

2.0.32 Changes since  2.0.31  (published 2026-10-13). New Features Custom shortcuts and saved setup Configure multiple shortcuts for common actions, including prompt undo, redo, and select all. You can also save your current model, reasoning, and setup choices as personal defaults. Selected terminal text can now be copied automatically with an opt-in auto-copy setting and clearer selection feedback. Simpler plan and artifact review Written plans can now be reviewed directly in the conversation. Linked visual artifacts are prepared automatically, remain available after resuming a session, and accept feedback through normal chat messages. Artifact previews can also be opened and exported directly from the CLI. Clearer context and model limits The CLI now shows active conversation context separately from total session token usage. Model limits are loaded from your configured inference service, and conversations compact before reaching the model’s safe request limit. The CLI also warns before switching models during an active task. More private remote sessions Remote sessions are now private to the user who started them. Fixes • Fixed Linear MCP connections not recovering after login and bug reports not resuming after authentication. • Shell tools now receive the correct shell and platform context, restoring commands that previously failed in some managed and Windows environments. • External filesystem access now requests consent consistently at tool boundaries without unnecessary prompts for worktree readiness checks. • Interrupted or cancelled runs now preserve prompts, task and tool progress, and committed file changes for session recovery. • Fixed inaccurate context usage reporting by using provider-reported token totals. • Improved terminal resizing and sidebar mouse interactions on Windows. • Keyboard shortcuts no longer trigger from key-release events. • Enter and Space now perform distinct actions, while Ctrl+S consistently saves default model and reasoning choices. • Custom inference API keys are now honoured in managed builds, including when loading model limits from your own inference endpoint. • Fixed archived sessions reappearing after concurrent session-index updates. • Restored research-only subagents in Auto mode. • Fixed secure secret input not being passed correctly to CLI tools. • Reasoning timers now continue correctly when moving between root views. • Local mobile control now requires explicit opt-in.