AgentGUI

Agents are great. As they work 24/7, how do we understand their actions and keep them on track? AgentGUI provides a local, open-source solution for observing and steering fleets of long-running agents.

AgentGUI is a locally hosted interface for many concurrent, long-running agent sessions. It visualizes agent trajectories, supports manual and automated steering — an autonomous manager audits each run against its task description and nudges it back on course — and integrates with both open-source and frontier agent frameworks. In the controlled user study reported in the preprint (July 2026), participants identified key elements of agent traces 38% faster. The paper was accepted at EMNLP 2026.