Skip to content

ChatSpeed v1.1.0

Choose a tag to compare

View all tags
@github-actions github-actions released this 03 Sep 03:57
v1.1.0
c5b59f2

🎉 ChatSpeed v1.1.0 Release Notes

Hello everyone! We are thrilled to announce the release of ChatSpeed v1.1.0. We've jumped directly from v1.0.4 to v1.1.0 because this version includes a series of milestone updates, especially regarding the proxy module's Tool Compatibility Mode and search engine support, which will significantly enhance the AI's capabilities and stability.

🚀 New Features

  1. Powerful Search Engine Integration

    • Now supports integration with major search engines via API keys, including Google Search API, Tavily API, and Serper API.
    • Additionally, built-in support for Bing, DuckDuckGo, So.com, and Sougou is included, providing you with diverse information sources.

  2. Web Content Scraping & Analysis

    • A new built-in web scraper has been added, capable of extracting text or Markdown content from a given URL, allowing the AI to fetch and analyze real-time web information.

  3. Flexible Prompt Injection

    • To better accommodate different models (especially those with weaker support for the system role), you can now choose to inject enhancement prompts into either the system role or the last user message, greatly improving compatibility.

✨ Enhancements

The core of this update is a deep refactoring and optimization of the proxy module's "Tool Compatibility Mode," aiming for a more stable, compatible, and fluid intelligent experience.

1. More Stable Tool Invocation

By refactoring the underlying tool compatibility mode, we have significantly improved the stability and accuracy of AI tool execution. Optimized tool definitions (with required parameters listed first and labeled (required)/(optional)) and stricter formatting rules provide a clear set of "guardrails" for the AI, better driving different models (even weaker ones) to execute tasks stably and accurately.

2. Customizable Cross-Model Compatibility

We've introduced a flexible prompt injection mechanism. You can now choose to inject enhancement prompts into the system or user message. This feature allows for optimization based on the characteristics of different models, solving issues where some models respond poorly to the system role, and significantly improving the performance of various models in clients like claude code.

3. Workflow Optimization via Prompt Guidance

We have greatly optimized the prompts to assist the AI in forming a more robust and less interruptible workflow. This includes guiding the AI in task planning (prioritizing tools like TodoWrite) and implementing intelligent fault tolerance (providing a detailed "correction guide" when tool call parsing fails), thereby significantly increasing the success rate of complex tasks.

🐞 Bug Fixes

  • Fixed a state management bug in the stream processor that caused tool call parsing to fail.
  • Fixed an issue where an XML parameter value would incorrectly become 0 or false if it could not be parsed as the specified type (e.g., number). It now safely falls back to preserving the original string, preventing data corruption.

We believe v1.1.0 will bring you a more powerful and reliable AI proxy experience. We have already felt the significant improvements ourselves while using the proxy module to connect clients like claude code, zed, and crush.

Thank you for your support!

🎉 ChatSpeed v1.1.0 发布日志

大家好！我们非常激动地发布 ChatSpeed v1.1.0 版本。本次我们从 v1.0.4 直接跳跃到 v1.1.0，因为它包含了一系列里程碑式的更新，特别是在代理模块的工具兼容模式和搜索引擎支持方面，这将极大地增强 AI 的能力和稳定性。

🚀 新增功能 (New Features)

  1. 强大的搜索引擎集成 (Powerful Search Engine Integration)

    • 现在支持通过 API 密钥接入主流搜索引擎，包括 Google Search API, Tavily API, 和 Serper API。
    • 同时，我们内置了对 Bing, DuckDuckGo, So.com, Sougou 等搜索引擎的直接支持，为您提供多样化的信息源。

  2. 网页内容抓取与分析 (Web Content Scraping & Analysis)

    • 新增了内置的网页抓取器（Scraper），可以从指定 URL 中提取文本或 Markdown 内容，让 AI 能够获取和分析实时网页信息。

  3. 灵活的提示词注入位置 (Flexible Prompt Injection)

    • 为了适配不同模型（特别是对 system 角色支持较弱的模型），现在您可以选择将增强提示词注入到 system 角色或最后一条 user 消息中，极大地提升了兼容性。

✨ 功能增强 (Enhancements)

本次更新的核心，是围绕代理模块的“工具兼容模式”进行的一系列深度优化，旨在提供更稳定、更兼容、更流畅的智能体验。

1. 更稳定的工具调用

通过对工具兼容模式的底层进行重构，我们极大地提升了 AI 执行工具的稳定性和准确性。优化的工具定义（必填项前置、(required)/(optional) 标注）和更严谨的格式规则，为 AI 提供了一套清晰的“护栏”，能更好地驱动不同模型（即使是能力稍弱的模型）稳定、准确地执行任务。

2. 可定制的跨模型兼容性

我们引入了灵活的提示词注入机制。现在，您可以选择将增强提示词注入到 system 或 user 消息中。这一功能使得我们可以针对不同模型的特性进行优化，解决了部分模型对 system 角色响应不佳的问题，显著改善了各类模型在 claude code 等客户端中的表现。

3. 通过提示词引导优化工作流

我们通过大幅优化提示词，来辅助 AI 形成更健壮、更不容易中断的工作流。这包括：通过引导 AI 进行任务规划（优先使用 TodoWrite 等工具），以及实现智能容错（在工具调用解析失败时提供详细的“纠错指南”），从而显著提升复杂任务的执行成功率。

🐞 Bug 修复 (Bug Fixes)

  • 修复了在处理流式响应时，因状态管理被意外破坏而导致的工具调用解析失败问题。
  • 修复了当 XML 参数值无法被解析为指定类型（如 number）时，会错误地变成 0 或 false 的问题，现在会安全地回退到保留原始字符串，避免了数据篡改。

我们相信，v1.1.0 将为您带来更强大、更可靠的 AI 代理体验。我们自己在使用代理模块接入 claude code、zed、crush 等客户端时，已经感受到了新版本带来的巨大提升。

感谢大家的支持！

Assets 17
Loading