Live Task Viewer
The console view that renders an agent's task event stream in real time as a task runs.
The live task viewer is the part of the Agenhood web console that renders a running agent task's event stream in real time, showing the agent's assistant messages, tool calls, tool results, and file changes as they happen, alongside a live token meter, iteration count, elapsed timer, and a control to cancel the task mid-flight.
How it works in Agenhood
When a task starts, the live task viewer opens a connection to that task's event stream and renders each incoming event as it arrives: text as it is generated, a tool call as it is issued, the tool's result once it returns, and any file written or modified in the agent's workspace. Counters for token usage, iteration count, and elapsed time update alongside the event feed, giving a running picture of cost and progress rather than only a final answer. Because the underlying event stream can be replayed from a disconnection point, reopening the live task viewer, for instance after a page reload, resumes the view where it left off instead of starting over.
Why it matters
Agent tasks can run for extended periods and take actions inside a sandboxed container that are not obvious from the final output alone. Watching tool calls and file changes as they happen makes it possible to catch a task going in the wrong direction and cancel it before it completes, rather than discovering the problem afterward. Real time observability into a long-running, tool-using process is a general need in agent infrastructure, not specific to Agenhood; a live, streaming view into an agent's actions is a common pattern for building trust in autonomous or semi-autonomous systems, whether the underlying agent is writing code, editing documents, or operating on external systems.
Related concepts
The live task viewer is a rendering layer on top of the task event stream; the event stream is the data, the viewer is the display. It is also distinct from the in-browser terminal, which gives direct interactive shell access into an agent's container rather than a read-only, structured view of a specific task's progress. A user can typically open both at once, watching a task unfold in the live task viewer while separately using the in-browser terminal to inspect the same container by hand.