This page covers what applies to all AI features in Astah: AI Chat Copilot, the MCP server and AI Code Reverse. To see what you can do with AI in Astah, with prompts and videos, visit AI Modeling in Astah.

On this page:

Data and privacy

  • Astah has no built-in AI and does not perform AI processing itself. AI features work only with an external AI service that you choose.
  • AI Chat Copilot, the MCP server and AI Code Reverse connect to that service with your own API key. Without an API key, they do not work. See API keys and costs.
  • With the MCP server, your AI client (for example Claude Code or GitHub Copilot) can also read and update your Astah model. The content it reads is sent to the AI service that client uses.
  • Content you ask the AI to work with, such as your prompt, model information and, for AI Code Reverse, source code, is sent from your computer to that AI service. Change Vision does not receive any of this content.
  • How your data is handled, including retention and use for training, is governed by the terms of the AI service you use. Check its policies before you start.
  • Follow your organization’s policy on confidential information. Do not send confidential or sensitive information unless your organization permits it.
  • In AI Chat Copilot, Allow Local File Access lets the AI read files in the folders you allow. Only allow folders that may be shared with the AI service.
    AI Chat Copilot settings in Astah with the Allow Local File Access option, which lets the AI read files in the folders you allow

API keys and costs

AI features in Astah use your own API key. You choose the AI service, and the AI provider bills you for usage based on its pricing.

  • OpenAI
  • Anthropic Claude
  • Google Gemini
  • OpenAI-compatible services
  • Azure OpenAI is supported in Astah Professional and Astah UML version 12 and later.
  • Costs grow with the amount of content sent and received. Large models, specifications or code bases use more tokens and cost more. Start with a small part, such as one package or module, and check the result before you run a large job.

Limitations

Keep these limitations in mind when you plan your work with AI in Astah.

Accuracy

  • AI-generated results may be incomplete or inaccurate. Always review and validate diagrams and models before you use them.
  • Results can vary, and unexpected behavior or problems can occur, depending on your prompt, the AI model and its settings, and the status of the AI service.
  • Back up your project before the AI edits your model. We strongly recommend saving a copy of the whole project first, so you can go back if the result is not what you want.

Token limits

AI services limit how many tokens they can process in one request. Sending or receiving very large specifications, models or code bases at once may:

  • reduce the quality of the response
  • cause timeouts or errors
  • make processing unstable

Split large work into smaller requests, for example one subsystem, one group of requirements, one package or one method at a time.

Guides for each AI feature

AI Chat CopilotChat with AI in the [Chat Copilot] tab to create, refine and review diagrams.Best for working inside AstahRead the guide → MCP serverLet AI clients such as Claude Code and GitHub Copilot read and update your Astah models.Best for working from your AI coding agentRead the guide → AI Code ReverseUse a dedicated menu to generate class and sequence diagrams from your source code.Best for understanding existing codeRead the guide →

For use cases with prompts and videos, see AI Modeling in Astah.