BEA
Planned scope: Multi-model AI agent with critic feedback, tool execution, web/RAG verification, and persistent memory.
Overview
Agent responses can require outside evidence, multiple steps, and context retained across tasks.
BEA explores a multi-model workflow that plans a task, uses tools, verifies evidence, and produces an answer with critic feedback.
Proposed workflow
- Plan the task.
- Execute tools and retrieve relevant material.
- Check the evidence with web and retrieval-assisted verification.
- Use critic feedback to revise the answer and retain relevant context.
Project status
A source repository, test suite, and evaluation results are pending publication.
Implementation status and deployment readiness remain unverified.