- 优化了流式UI断触、用户中断不渲染（但是仍然保存到上下文，只是不美观）
- 优化了超长对话、每轮对话快结束大量UI生成的渲染卡顿问题
- 优化了对话时流式UI效果
- 模型卡片添加上下文窗口大小和最大输出配置（不配置则默认128K、12.8K）
- 增强了Authropic协议的支持
- 添加了Linux和MacOS的webview渲染H5代码，但是未做验证
- 除此之外还优化了其他卡顿问题和UI
- Optimized the streaming UI to handle touch interruptions and prevent rendering when the user interrupts (though the data is still preserved in the context, even if the visual result is less polished).
- Optimized performance to resolve UI rendering lag during very long conversations and when large amounts of UI content are generated near the end of a turn.
- Refined the streaming UI experience during conversations.
- Added configuration options for context window size and maximum output tokens to model cards (defaulting to 128K and 12.8K if not specified).
- Enhanced support for the Anthropic protocol.
- Added WebView-based H5 rendering code for Linux and macOS (though not yet verified).
- Implemented various other UI and performance optimizations to reduce lag.
- 优化了长对话思考时途中可能产生卡顿的问题.
- 优化了"软失效过滤"，避免过时记忆和新记忆同权重混在检索结果里.
- 编排多Agent对大量文档进行并行理解的命令.
- 可控的agentloop模式，从思考，编写，测试，验证的循环流水线.
- 技能现在可以批量导入.
- 优化了代码高亮文本.
- 使用archify对对话进行流程图总结输出.
- 更为可靠的版本工作流.
- 知识库功能.
- 优化了MCP服务器的载入和启动配置.
- 优化了Agent自主生成文件默认在工作目录下.
- Optimized the issue of potential stuttering during long conversations.
- Optimized the "soft failure filter" to prevent outdated and new memories from being mixed with the same weight in search results.
- Orchestrated commands for multi-agent parallel understanding of large amounts of documents.
- A controllable agentloop model, a pipeline of thought, writing, testing, and verification.
- Skills can now be imported in batches.
- Optimized code highlighting.
- Use archify to summarize and output flowcharts of conversations.
- More reliable version workflows.
- Knowledge base functionality.
- Optimized MCP server loading and startup configuration.
- Optimized the default location of Agent-generated files in the working directory.
- 提供了在线访问Agent的功能
- 优化了上下文压缩
- 在工作目录下创建、使用临时技能进行临时性自我进化
- 命令功能，比如说创建全局技能并且自己导入
- Provides online access to the Agent.
- Optimized context compression.
- Create and use temporary skills for temporary self-evolution within the working directory.
- Command functionality, such as creating global skills and importing them automatically.
- 优化了部分实现、增强了工具链
- 在模型输出html5的时候可以使用webview2预览了
- Optimized some implementation details and enhanced the toolchain
- WebView2 can now be used for previewing when outputting HTML5 models
- 添加了字体设置
- 内置了截图工具
- 检测到大量模型时可以输入快速匹配
- 添加了宠物功能，宠物作为全局观察者，可以通过右键内容唤醒进行交互
- Added font settings
- Built-in screenshot tool
- Allows for quick matching when a large number of models are detected
- A pet feature has been added. Pets act as global observers and can be activated and interacted with by right-clicking on content.
发布！支持MCP、Skill、Search、Memory、多Agents协同办公等等等
Released! Supports MCP, Skill, Search, Memory, multi-agent collaborative work, and more.