The EnCompass system runs AI agent programs by backtracking and making several attempts, finding an LLM’s best set of outputs. It lets programmers easily experiment with different search strategies so they can work with AI agents more efficiently. The work was led by researchers at MIT CSAIL and Asari AI.

MIT is a renowned institution for education and research, offering insights into science, engineering, and technology. Through publications, research papers, and academic programs, MIT's platform provides insights into  research, innovation, and education in various fields. Students, researchers, and technology enthusiasts can learn about MIT's contributions to science and technology and explore opportunities for academic and professional development.

MIT News

MIT CSAIL and Asari AI researchers developed EnCompass, a framework that helps AI agents using large language models backtrack and retry when mistakes occur. The system separates search strategy from workflow logic, allowing programmers to experiment with different approaches like Monte Carlo tree search or beam search without extensive code changes. In testing, EnCompass reduced coding effort by up to 82% and improved accuracy by 15-40% on code translation tasks. The framework enables developers to build more reliable AI agents by automatically handling LLM errors and exploring multiple execution paths to find optimal solutions.

Helping AI agents search to get the best results out of large language models