简体中文 | English

ChatSpeed

Forged in Rust, this open-source, high-performance AI assistant is your powerful programming companion and smart desktop hub.

Claude Code Integration Demo

🌟 What can Chatspeed do?

  • ⚡ Tool-Driven Workflow Engine: Built for coding and long-running multi-step work. Chatspeed uses a tool-driven Plan -> Execute -> Final Review loop to reduce task interruption and push work toward completion instead of stopping at partial analysis.
  • 🤖 Main Agent + Child Agents: Split work across a parent agent and isolated sub-agents for parallel subtasks, review passes, and controlled delegation.
  • 🧠 Memory + Skills + MCP: Combine persistent memory, reusable skills, and MCP-based tool expansion into one execution system instead of scattering capabilities across separate products.
  • 🧩 Multi-Model by Design: Mix planning, execution, utility, and vision models. Supports OpenAI-compatible, Gemini, Claude, and Ollama models, while CCProxy extends tool access to models without native tool calling.
  • 💼 Multi-functional Desktop Assistant: You can also use it for translation, mind maps, flowcharts, daily conversations, and other desktop tasks, quickly summoned with ALT+Z.
  • 🔌 Connect Any Dev Tool: Beyond just Claude Code, you can also connect models to almost any major AI development tool, including Gemini CLI, Cline, Roo Code, and Zed.
  • 💰 Use Claude Code for Free: As a best practice, we provide a detailed tutorial on how to use Claude Code for free.
  • 🚀 MCP Hub: Chatspeed's MCP proxy provides its own WebSearch and WebFetch tools, along with any MCP tools you've installed, to external clients via the stable Streamable HTTP protocol (SSE protocol has been removed from v2.0). Learn how to centrally manage MCP.

[!CAUTION]

⚡ Workflow-First Philosophy

Chatspeed's workflow system is built around one practical belief: serious coding work should be tool-driven, stateful, and completion-oriented.

Many AI coding tools fail in the middle of real work because they drift into chat, lose task state, or stop after partial reasoning. Chatspeed is designed to push in the opposite direction:

  • Tool-driven execution instead of chat-only continuation
  • Structured Plan -> Execute -> Final Review instead of one-shot answers
  • Explicit approvals, checkpoints, and recovery instead of hidden state
  • High completion pressure for real tasks, not just good-looking intermediate output

This is the main distinction of Chatspeed's workflow engine: it is not trying to be another generic chat wrapper. It is trying to be a reliable task completion system for real engineering work.

🚀 Core Engine: CCProxy

Chatspeed's power is driven by its core proxy engine, CCProxy. It's a universal adapter built with Rust, focused on:

  1. Protocol Conversion: Seamlessly convert between major protocols like OpenAI-compatible, Claude, Gemini, and Ollama.
  2. Capability Expansion: Expands the capabilities of models that do not natively support tool calling through a tool compatibility mode.
  3. Model Enhancement: CCProxy's prompt injection feature effectively improves the performance of non-Claude models when connected to Claude Code, transforming them from conversational models into professional code assistants.
  4. Reducing User Burden: The tool compatibility mode frees users from worrying about whether a model supports native tool calling, significantly lowering the barrier to entry and mental load.
  5. Security Isolation: CCProxy's keys effectively isolate clients from direct access to AI provider keys, enhancing key security.
  6. Group Management: Supports a proxy grouping feature to restrict client access to specific model groups.
  7. Load Balancing: Effectively mitigates model rate-limiting issues by globally rotating through all configured provider keys.
  8. Simplified Workflow: Say goodbye to repeatedly configuring tools in different IDEs with a unified MCP entry point.

⚡ Workflow Engine

Chatspeed introduces a tool-driven Workflow Engine since v2.0, purpose-built for complex multi-step tasks like coding, debugging, refactoring, and documentation work. Key capabilities include:

  • Plan -> Execute -> Final Review Main Loop: Planning is not just a prompt style. It is an explicit workflow stage with approval and transition into implementation.
  • Main Agent + Child-Agent Orchestration: Delegate parallel or isolated work to sub-agents while preserving parent workflow control.
  • Skills Integration: Reuse structured skills as part of workflow execution rather than bolting them on as separate prompts.
  • MCP Expansion: Extend the workflow through MCP tools and use the same MCP ecosystem across Chatspeed and external clients.
  • Persistent Memory: Keep both project memory and global memory across sessions.
  • Multi-Model Workflow Routing: Assign different models to planning, execution, utility, and vision roles.
  • Broad Model Compatibility: Supports OpenAI-compatible, Gemini, Claude, and Ollama protocol families. CCProxy can bridge some models that do not support native tool calling.
  • Approval, Safety, and Review Controls: Auto-approve, shell policies, PathGuard, AI-assisted review, and explicit terminal completion tools.
  • Recovery and Resume: Workflow state, approvals, waiting, and recovery are designed as first-class runtime concerns instead of best-effort UI behavior.
  • High Context-Cache Efficiency: In real coding tests with deepseek-v4, context-cache hit rate is typically around 90%+, and in some programming scenarios reached as high as 98%.
  • Dedicated Workflow UI: Approval dialogs, file diffs, task ledger, status panel, multi-workflow switching, and execution-state feedback are built in.

If you care about whether an AI coding system can finish multi-step work rather than merely talk about it, this is the part of Chatspeed to pay attention to.

Learn more about the Workflow Engine.

Installation

Windows

  1. Download the .exe installer from the Releases page.
  2. Run the installer and follow the on-screen instructions.
  3. You may see a Windows SmartScreen warning. Click "More info," then "Run anyway" to proceed.

macOS

Download the .dmg file from the Releases page. For Apple Silicon, choose the file with the _aarch64.dmg suffix; for Intel chips, choose the _x86.dmg suffix.

Important Note: On recent versions of macOS, the Gatekeeper security feature may prevent the app from running and show a message that the file is "damaged". This is because the application has not yet been notarized by Apple.

Please use the following terminal command to resolve this issue:

  1. Drag the .app file from the mounted .dmg image to your "Applications" folder.

  2. Open the Terminal app.

  3. Execute the following command (you may need to enter your system password):

    sudo xattr -cr /Applications/Chatspeed.app

  4. After the command executes successfully, you can open the application normally.

Linux

  1. Download the .AppImage, .deb, or .rpm file from the Releases page.
  2. For .AppImage files, first grant execute permissions (chmod +x Chatspeed*.AppImage), then run it directly.
  3. For .deb files, use your package manager to install, or install via the command sudo dpkg -i Chatspeed*.deb.
  4. For .rpm files, use your package manager to install, or install via the command sudo rpm -ivh Chatspeed*.rpm.

📚 Learn More

We highly recommend starting with our Official Documentation Website for the best reading and learning experience.