See Your Agents In Action

Supports 100%local deployment and inference

Agents are great. As they work 24/7, how can we best understand their traces, and keep them on track?
AgentGUI provides a living pixel-art interface to observe and orchestrate teams of AI agents in real time.

View on GitHub
$ git clone https://github.com/eth-medical-ai-lab/agent-gui
What it does

See and manage your team of agents in action.

AgentGUI helps humans to understand actions of autonomous agents. Inspect reasoning, tool calls, artifacts and outputs behind each step, all in one view. An autonomous manager keeps long-running jobs on track.

Seamless integration with Hermes agent

AgentGUI automatically imports your Hermes agents. Assign an agent into action with a single click.

Audit agents automatically as they execute long-running tasks

An autonomous manager watches agent traces, and audits them against a modifiable task description.

Designed with security, privacy, and local performance in mind

Supports fully local inference solutions. Agent terminal actions are confined within sandboxes. Select minimal tool sets for each agent at run time for faster inference.

Assemble your team

Customize agent configurations, persona, and files shared across a team of agents. Also theme the workspace, pick agent avatars, code themes.

Demo

AgentGUI in Action

01

Kick-start an analysis with a single click

loop

Drop a file or a prompt staright onto the agent desk to get started.

02

Autonomous auditing and steering

loop

The manager patrols automatically audits an agent's work against the TASK.md, marks it as solved or steers the task back on course.

03

Visualizing and inspecting long-horizon tasks

loop

View the long history and artifacts as a coder agent grinds through a multi-step build.

04

Save and load trajectory for review

loop

Load a saved trajectory to review an agent's actions and results.

A closer look

Workplace details, feature by feature

The floor

A living office for your agents

Monitor all details of your agents' activities in one place.

  • Monitor run details of agents on a dashboard.
  • Organize agents by task topic and create shareable directories.
  • Spin up agents with targeted memory for the same task.
agentgui · the office floor
AgentGUI office floor
At the desk

Every view an agent run needs

Open any desk to inspect what happened — from a high-level run breakdown down to raw terminal output and full chat history.

  • Desk history — reload saved trajectories for review.
  • Run breakdown — see reasoning, tools, and idle time at a glance.
  • Debug terminal — watch sandbox shell I/O live.
  • Chat history — scroll the full agent conversation.
agentgui · desk history
Saved desk history view

Saved desk history — load a prior session and replay its trajectory.

agentgui · run breakdown
Run breakdown overview

Run breakdown overview — time-resolved view of reasoning, tools, and idle stretches.

agentgui · debug terminal
Debug terminal view

Debug terminal — inspect sandbox shell commands and output.

agentgui · chat history
Chat history view

Chat history — read the full back-and-forth with the agent.

Deployment

Arrange and manage your agents

Create customized Hermes profiles, and select at run time minimal tool scope for efficiency.

  • Fully local app deployment available — no cloud required.
  • Select run time tool access per agent, so each gets only what it needs.
agentgui · deployment
AgentGUI deployment — local run with scoped tool access
Oversight

A manager that keeps watch

An autonomous manager patrols the floor, audits each agent's work against its task, and intervenes when results miss the mark. Long-running jobs stay on track without requiring manual intervention.

  • Audits output against the original task spec.
  • Intervenes with another pass when checks fail.
  • Tunable idle threshold to nudge at the right cadence.
agentgui · manager audit