Highlights
- show Recent Models in /model
- improve plan mode like claude-code
- add providers codex and qwen with oauth, and remove antigravity provider
- support acp
- upgrade ai to 6 and fix kimi k2.5 and deepseek r1 problems
- improve prompt caching
- improve thinking config
- improve the way to custom provider and models
- improve auto compact strategy
- log llm request and response to request jsonl file
- truncate large tool output and use file as ref
- support kimi k2.5 model
- fix llm cancel don't stop immediately
- refact run command
- refact commit command for simplify
What's Changed
- fix: add fallback beep for notification sound errors by @sorrycc in #689
- fix: allow model command to bypass initialization error by @sorrycc in #690
- deps: update ink from 6.5.1 to 6.6.0 by @xierenyuan in #691
- refactor: remove unused tool approval logic by @xierenyuan in #696
- feat: optimize slash command UI with fixed width and truncation by @xierenyuan in #695
- fix: record tool results for remaining tools on denial by @sorrycc in #698
- feat: add skills option to sdk session options by @sorrycc in #694
- chore: remove biome code actions on save by @xierenyuan in #702
- feat: add repair to safeParseJson by @xierenyuan in #700
- fix: win32 spawn npx enoent error by @sorrycc in #705
- fix: ensure executable permissions for bundled ripgrep binary by @sorrycc in #706
- feat(at): ignore directory @mentions by @xierenyuan in #701
- refactor: reorganize tool resolution order in resolveTools by @xierenyuan in #708
- refactor(run): rewrite run command with readline interface by @sorrycc in #710
- feat: add responses model type support by @sorrycc in #711
- refactor: rename createModelType to apiFormat by @sorrycc in #712
- feat: add content output limits to grep tool by @xierenyuan in #707
- fix: handle terminal focus events globally by @sorrycc in #713
- feat: add external signal synchronization for loop cancellation by @xierenyuan in #714
- fix: improve tool argument display truncation and formatting by @xierenyuan in #715
- feat: add truncation plugin for automatic tool output management by @xierenyuan in #718
- refactor: support language-specific compact system prompts by @xierenyuan in #720
- refactor: move ProviderConfig export to fix duplicate export by @xierenyuan in #721
- feat: upgrade ai-sdk to v3 and update language model types by @sorrycc in #722
- feat: implement auto compression strategy optimization by @xierenyuan in #719
- fix: handle AI SDK v6 usage format with nested token objects by @xierenyuan in #724
- refactor: move providers to modular structure with pipeline by @sorrycc in #725
- fix: add fallback for undefined model in provider normalization by @xierenyuan in #726
- refactor: move thinking config to model variants by @sorrycc in #727
- feat: enhance prompt caching and api format support by @sorrycc in #728
- feat: add onRequest/onResponse HTTP logging hooks by @sorrycc in #729
- refactor: move CLAUDE.md rule loading into conditional block by @xierenyuan in #730
- feat: acp support by @suxin2017 in #717
- feat: implement complete plan mode with 5-phase workflow by @xierenyuan in #709
Full Changelog: 0.25.0...0.26.0