feat: migrate to AI SDK v5.0.0-beta Agent abstraction

[miurla]

Summary

• Migrated from traditional streamText API to new Agent abstraction introduced in AI SDK v5.0.0-beta.8
• Implemented sequential agent execution pattern for research and related questions generation
• Fixed message handling and data structure issues

Changes

• Agent Migration:

  • Updated researcher function to return an Agent instance
  • Updated generateRelatedQuestions to use Agent pattern
  • Implemented proper activeTools and stopWhen configurations

• Sequential Execution:

  • Research agent runs first with sendFinish: false
  • Related questions agent runs after research completes with sendStart: false
  • Proper message merging for saving complete responses

• Bug Fixes:

  • Fixed API route to handle AI SDK v5 messages array format
  • Fixed related questions component to handle both array and object data formats
  • Improved error handling and abort signal support

• Improvements:

  • Enhanced related questions prompt to avoid repetition and generate unique follow-ups
  • Added proper TypeScript types and removed debug code
  • Maintained backward compatibility with existing UI components

Test Plan

• Test basic chat functionality
• Verify research results are displayed correctly
• Confirm related questions are generated and displayed
• Test error handling when agents fail
• Verify message saving works correctly
• Run linting and type checking
• Test production build

Breaking Changes

• Requires AI SDK v5.0.0-beta or higher
• Agent abstraction is experimental and may change in future releases

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Comments	Updated (UTC)
morphic	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	Jul 26, 2025 7:53am

[Copilot]

Pull Request Overview

This PR migrates the chat system from the traditional streamText API to the new Agent abstraction introduced in AI SDK v5.0.0-beta.8, implementing a sequential execution pattern for research and related questions generation.

• Migration from streamText API to Agent-based architecture with proper tool configuration
• Implementation of sequential agent execution where research runs first, followed by related questions generation
• Enhanced message handling to support AI SDK v5 message formats and improved error handling

Reviewed Changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
lib/streaming/create-chat-stream-response.ts	Implements sequential agent execution pattern with proper message merging and error handling
lib/agents/researcher.ts	Converts researcher function to return Agent instance with updated tool configuration
lib/agents/generate-related-questions.ts	Migrates to Agent pattern with improved prompt to avoid question repetition
components/related-questions.tsx	Adds backward compatibility for both array and object data formats
app/api/chat/route.ts	Updates API route to handle AI SDK v5 messages array format

Comments suppressed due to low confidence (1)

lib/agents/researcher.ts:60

• The tool name has been changed from 'ask_question' to 'askQuestion'. Ensure this naming change is consistent with tool usage elsewhere in the codebase, as the snake_case to camelCase change could break existing references.

        askQuestion: askQuestionTool

[Copilot]

[nitpick] The magic number 10 for maximum steps should be extracted to a constant for better maintainability. Consider defining MAX_SEARCH_STEPS = 10 to make this configurable and self-documenting.

Suggested change

	stopWhen: searchMode ? stepCountIs(10) : stepCountIs(1)
	stopWhen: searchMode ? stepCountIs(MAX_SEARCH_STEPS) : stepCountIs(1)