Skip to main content

Product Overview

AgentHub is a control plane for AI-agent work. It keeps agent sessions alive, gives operators one place to inspect and steer them, and adds structured history instead of relying on raw terminal scrollback alone.

Core Idea​

AgentHub is not just a prompt box for one terminal session. It combines:

  • long-lived agent runtimes
  • a web control surface for start, stop, reconnect, and review
  • structured ACP output rendering
  • optional Team workflows with leader and worker roles
  • optional remote execution nodes for multi-machine rollout

The important shift is that AgentHub treats the browser as a control surface, not as the place where the agent actually lives. The runtime, history, and coordination state stay on the backend.

Why Teams Use It​

AgentHub fits especially well when engineering work needs more than a single interactive shell:

  • one operator may need to reconnect to the same long-lived session many times
  • multiple people may need to inspect the same run or Team state
  • implementation and review may need to be split across multiple agents
  • remote execution may need to preserve the same actor mailbox model as local execution

When AgentHub Fits Well​

AgentHub is a strong fit when you need one or more of these:

  • browser disconnect should not stop the task
  • multiple operators need a shared control surface
  • output should stay replayable and auditable
  • work should run in isolated worktrees by default
  • one machine is no longer enough for all agent execution

Choose The Right Workflow​

  • Agents: best for one operator driving one agent session directly.
  • Teams: best when planning, implementation, and review should be split across multiple members.
  • Agent Nodes: best when execution must move to other machines but the main control plane should stay central.
  • OpenAPI: best when AgentHub must be integrated into scripts, CI, or other internal tooling.

What Makes AgentHub Distinct​

Three properties tend to matter together:

  • long-lived runtime control: browser refreshes and disconnects do not end the work
  • structured ACP history: plans, tools, output, and debug data stay reviewable after the run
  • actor p2p execution model: remote execution keeps the same actor and mailbox control path instead of introducing a separate orchestration system

What AgentHub Persists​

AgentHub keeps operational state in ~/.agenthub/ by default, including:

  • SQLite-backed runtime and history state
  • agent configuration
  • Team state
  • audit and operational records

This is why reconnect, history replay, and deployment operations can happen without treating the browser as the source of truth.

Next Reading​