MGX Docs

Appearance

MGX

Tips & tricks

How Can I Optimize the Use of Credits ​

MGX is a magical, multi-agent, natural-language programming tool. For first-time users, we highly recommend reviewing the following materials to avoid the rapid consumption of credit.

Start with the Basics ​

  • Begin with Simple Requirements:When you start using MGX, begin with small-scale function tests. This allows you to gradually understand MGX's functions and discover relevant use cases.

  • Leverage Version Control for Step-by-Step Improvements:

    • Start with a Minimum Viable Product (MVP) and gradually iterate on the project.
    • By activating version summaries through the remix function, you can effectively prevent rapid credit consumption that may result from increasing usage scales and memory volumes.

Make Prudent Model Selections ​

  • Basic Tasks:For basic tasks such as data analysis, simple programming, and general Q&A, DeepSeek-chat is a good choice due to its lower cost.
  • Complex Scenarios:For complex MGX scenarios where optimal results are desired, the Anthropic Claude Sonnet model is recommended, although it has a higher Credit consumption.
  • Future Outlook:An Auto Mode will be introduced in the future. It will seamlessly integrate the capabilities of various Large Language Models (LLMs), balancing performance and Credit efficiency.

Summarize and Initiate New Projects in a Timely Manner ​

  • Control Conversation Length:Credit consumption increases with the length of conversations. To minimize usage：
    • Create separate chats for basic, non-project-related Q&A.
    • After clarifying requirements, summarize the key points and start a new chat for implementation.
    • Once the MVP is complete, summarize its details in a README.md file, archive the code (on GitHub, GitLab, or locally), and begin a new conversation for incremental module development. Start with the MVP and iteratively develop the project. By generating a version summary through the remix function, you can effectively prevent rapid credit consumption as the usage scale and memory volume grow.

Be Exact with Your Requirements ​

  • Plan Tasks Thoroughly:Plan incremental tasks in detail and describe them clearly.
  • Consolidate Modification Requests:When modifying the same module, consolidate all changes into a single request to avoid repeated clarifications that increase credit consumption.
  • Use Technical Means:Technical users can use "#" to specify files or highlight particular content for editing.
  • Keep Descriptions Concise:Keep descriptions concise and avoid including irrelevant details.

Optimize Chat Strategy ​

  • Avoid Repeating Issues:Do not revisit the same issue multiple times in conversations.
  • Break Down Complex Tasks:For complex tasks, manually break them down and guide the agents step by step.
  • Limit Interaction Rounds:Limit the number of interaction rounds within a single conversation or project.
  • Steer Clear of Long-Form Tasks:Tasks like blog writing can result in excessive credit use and should be avoided.

Develop Best Practices ​

  • Utilize Scenario Templates:Leverage scenario templates for quick project setups.
  • Use File Interactions Efficiently:Have David output formatted files (e.g., JSON, Markdown), and allow Alex to build upon these files in new conversations, reducing redundant processing.
  • Module - based Handling of Large Projects:For large projects, divide them into smaller modules, complete each in separate conversations, and merge them using GitHub, GitLab, or local tools.

Remix Function ​

  • Advantages of Remix:Click "Remix this version" to launch new dev chats from past versions. Remix helps retain context during development, tracks changes via version summaries, and optimizes resource use. It prevents rapid credit loss from expanded usage or memory growth.
  • Ways to Remix:
    • Directly click the remix icon within the conversation area to branch off the current version.
    • Select a specific version from the interactive progress bar to remix from any checkpoint.
    • Navigate to the "Versions" section at the bottom of the page, explore past iterations, and remix your preferred version.