Task Event Stream
The ordered sequence of structured events an Agenhood agent emits while a task is running.
The task event stream is the ordered sequence of structured events that an Agenhood agent emits while an agent task is running, including message chunks, tool calls, tool results, file changes, errors, and a final completion event, delivered over Server-Sent Events, or SSE.
How it works in Agenhood
As an agent works through a task, calling tools and editing files inside its sandboxed container, Agenhood emits a stream of discrete events describing what is happening: a chunk of the assistant's message, a tool call being made, the result of that call, a change to a file in the workspace, an error, and eventually a done event marking task completion. These events are ordered and can be consumed by any client over SSE, not only Agenhood's own console. If a client disconnects, the stream can be replayed from the point of disconnection, so a reconnecting client resumes exactly where it left off rather than losing or duplicating events.
Task Event Stream vs Live Task Viewer
The task event stream is the underlying data feed: an ordered, replayable sequence of structured events produced by the platform. The live task viewer is one consumer of that feed, a user interface in the web console that renders the events as they arrive, adding presentation such as a token meter, iteration count, and elapsed timer. Any other client, such as a script or an integration, could consume the same task event stream over the API without ever touching the live task viewer.
Why it matters
Structuring task progress as a well defined event stream, rather than a single blocking response, lets observers see what an agent is doing while it is still working, rather than only after it finishes. Streaming structured events over SSE with resumable playback is a general pattern used across chat and agent systems for showing incremental progress; Agenhood applies it specifically to tool calls, file changes, and task completion inside its sandboxed agents. The same stream also doubles as a record of what happened during a task after it finishes, since every message chunk, tool call, and file change that occurred while the agent worked remains available for later review.