You need credits to interact with agents. To avoid wasting credits, start with a simple prompt, select the right model, plan your projects in advance, and utilize the Remix function effectively. Additionally, there are strategies for chatting with agents. Let's go through them!
Start with the Basics
A simple beginning: Try lightweight functions first, such as "sheets", "docs", or "Engineer Mode". This helps you understand MGX's functions and gradually discover relevant use cases.
Organized iteration:
Build a Minimal Viable Product (MVP) first and iterate it step by step.
Remix helps you to activate version summaries. It prevents rapid credit consumption.
Select Right LLMs
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 cases: For complex cases requiring optimal results, the Anthropic Claude Sonnet model is recommended. And it also consumes more credits.
Future Outlook: An Auto Mode will be introduced in the future. It will integrate the capabilities of various LLMs, balancing performance and Credit efficiency.
Control Conversation Length
Credit consumption increases with the length of conversations. Try to keep the conversation shorter, and it helps save credits. You minimize it by:
Create separate chats for basic, non-project-related Q&A.
Once the requirements are clear, summarize the key points and start a new chat to build.
Summarize the details of the MVP in a README.md file, archive the code (on GitHub, GitLab, or locally). Then begin a new chat with agents to build step by step, adding one feature/module at a time.
Keep requirements precise
Plan Thoroughly: Plan projects step by step and describe them clearly, including all necessary details.
Summarize all necessary changes: When modifying the same module, summarize all changes in a single request to avoid repetition.
Use Technical Shortcut: Technical users can use "#" to specify files or highlight particular content for editing.
Chat strategically
Avoid Repeating: Refrain from revisiting the same issue multiple times in conversations.
Break Down Complexity: For complex tasks, break them down manually and guide agents step by step.
Limit Interaction Rounds: Restrict the number of interaction rounds within a single conversation or project.
Avoid Long-Form Tasks: Try to avoid lengthy writing tasks (such as full blog posts), as they consume a large amount of credits.
Develop Best Practices
Use Templates: Use templates of user cases for quick project setups.
Use File Interactions Efficiently: For instance, have David output formatted files (e.g., JSON, Markdown), and allow Alex to build upon these files in new chats, thereby reducing extra processing.
Develop in Modules: For large projects, divide them into smaller modules, complete each in separate chats, and integrate them using GitHub, GitLab, or local tools.
Remix
You can also remix the past versions of projects for development.
Three ways of Remix:
Directly click the remix icon in the chatbox
Select a specific version from the interactive progress bar to remix from any checkpoint.
Navigate to the "Versions" section at the top of the page, explore past iterations, and remix your preferred version.
For more information, see: Remix.