功能优化与新增
📝 文本与编辑优化
- 富文本编辑器支持：提升提示词编辑体验，支持更丰富的格式调整。
- 拖拽输入优化：支持从其他软件直接拖拽文本至输入框，简化内容输入流程。
- 参数调节增强：新增 Top-P 和 Temperature 开关设置，提供更灵活的模型调控选项。
- 翻译任务后台执行：翻译任务支持后台运行，提升多任务处理效率。
🛠️ 模型与推理改进
- 新模型支持：新增 Qwen-MT、Qwen3235BA22Bthinking、GLM 4.5 系列和 sonar-deep-research 模型，扩展推理能力。
- 推理稳定性提升：修复部分模型思考内容无法输出的问题，确保推理结果完整。
- Mistral 模型修复：解决 Mistral 模型无法使用的问题，恢复其推理功能。
📂 文件与备份管理
- 备份目录优化：支持 相对路径输入，提升备份配置灵活性。
- 数据导出调整：新增 引用内容导出开关，提供更精细的导出控制。
⚡ 性能与稳定性修复
- 文本流完整性：修复文本流末尾文字丢失问题，确保输出内容完整。
- 内存泄漏修复：优化代码逻辑，解决内存泄漏问题，提升运行稳定性。
- 嵌入模型简化：降低嵌入模型配置复杂度，提高易用性。
⏳ 长期运行支持
- MCP Tool 长时间运行：增强 MCP 工具的稳定性，支持长时间任务执行。
What's Changed
- fix(AiProvider): remove unnecessary middleware removal logic for… by @DeJeune in #8437
- Feat/vertex-claude-support by @DeJeune in #7564
- fix(mcp-tools): enhance tool lookup logic to support partial matches by @DeJeune in #8473
- chore(deps): update vite to rolldown-vite by @beyondkmp in #8460
- refactor(ThinkChunkMiddleware): remove reasoning check for improved logic clarity by @DeJeune in #8505
- chore: ignore qwen-code settings by @alephpiece in #8509
- chore: Update channel improve by @beyondkmp in #8501
- fix: aihubmix provider generate image logic by @DDU1222 in #8478
- refactor(MessageEditor): lower minrows to 1, add padding by @alephpiece in #8434
- feat: use code editor in prompt settings by @alephpiece in #8456
- fix/mcp-bug by @DeJeune in #8518
- feat: support Relative Path Input for Backup Directory by @beyondkmp in #8471
- feat(MCPSettings): enhance JSON loading to inherit all MCPServer fields by @vaayne in #8498
- chore(i18n): forced nested structure to support i18n ally by @EurFelux in #8457
- refactor(useSmoothStream): optimize chunk handling and state management by @MyPrototypeWhat in #8514
- fix: call tool bug by @DeJeune in #8526
- fix(i18n): standardize i18n usage by @EurFelux in #8525
- fix: set consolelog eslint to error when in prci lint check by @0xfullex in #8532
- refactor(Knowledge): simplify dimension settings, support base migration by @alephpiece in #8315
- Feat/long run mcp by @DeJeune in #8499
- fix: #8531 by @Pleasurecruise in #8535
- feat(translate): add Ukrainian by @EurFelux in #8528
- feat(models): add support for new Qwen model and update effort ratio by @DeJeune in #8537
- fix(i18n): add i18n for data restore by @EurFelux in #8533
- feat(export): citation export control by @GeorgeDong32 in #8519
- Fix/qwen-mt by @DeJeune in #8527
- feat(mcp): add shouldConfig flag and i18n support for builtin MCPServer description by @EurFelux in #8440
- refactor(OpenAIApiClient, models, ThinkingButton): streamline reasoning model checks and enhance support for Perplexity models by @DeJeune in #8487
- feat(models): add Grok 4 model and update reasoning regex by @GeorgeDong32 in #8545
- feat(assistants): place copied assistant after the original one by @EurFelux in #8557
- perf(i18n): improve performance when getting i18n text by @EurFelux in #8548
- fix: pangu-pro-moe not reasoning by @Pleasurecruise in #8572
- fix(ModelEditContent): preserve model price settings by @EurFelux in #8560
- feat(contentsearch): optimize searchbar UI by @GeorgeDong32 in #8556
- feat(inputbar): drop text into inputbar by @EurFelux in #8579
- fix: add isPathInside functionality to check path relationships by @beyondkmp in #8590
- feat: add Top-P parameter toggle with default enabled state and improved UI styling by @lihqi in #8137
- feat(installer): add architecture compatibility check to NSIS installer by @beyondkmp in #8587
- fix(LocalBackup): streamline local backup relative directory handling by @beyondkmp in #8595
- feat(models): add support for Zhipu model and enhance reasoning checks by @DeJeune in #8609
- fix(WebSearchSettings): responsive table layout by @EurFelux in #8606
- fix(think-tool): update prompt handling logic and improve comments by @DeJeune in #8538
- fix: show something reasonable when missing embedding model by @EurFelux in #8600
- feat: improve builtin mcp sever display style by @kovsu in #8470
- feat(translate): support background execution of translation tasks by @EurFelux in #7892
- fix: no
/no_thinkfor qwen3 anymore if provider is dashscope by @EurFelux in #8616 - refactor(ModelEditContent): enhance model capabilities management and… by @DeJeune in #8562
- fix: reset pyodide globals between runs by @alephpiece in #8620
- fix/export-roletext-level by @GeorgeDong32 in #8634
- fix: change jschardet to chardet by @Pleasurecruise in #8577
- perf: part of memory leak by @EurFelux in #8619
- integrate AWS Bedrock API by @silenceboychen in #8383
New Contributors
Full Changelog: v1.5.3...v1.5.4-rc.1