> ## Documentation Index
> Fetch the complete documentation index at: https://docs.tryblend.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Running a single node

> Run one node on its own to test it, with its results kept right on the node.

In the [editor](/workflow/overview), supported server-side nodes can run **by themselves** without running the whole graph. It's the fastest way to test a model or server operation before committing to a full run.

When you run a single node, Blend AI resolves just that node's inputs from whatever is wired upstream (or already produced), runs it, and streams the result back live.

## Supported nodes and modes

| Node or mode                                                          | Single-node run                                         |
| --------------------------------------------------------------------- | ------------------------------------------------------- |
| LLM **Agent**                                                         | ✅ Supported                                             |
| LLM **Sub-agent**                                                     | ❌ Connect it to a parent Agent                          |
| Image Gen                                                             | ✅ Supported                                             |
| Video Gen                                                             | ✅ Supported                                             |
| Tool **Standalone**                                                   | ✅ Supported                                             |
| Tool **Agent tool**                                                   | ❌ Local tool-list wiring only                           |
| Utility **Extract Audio**                                             | ✅ Supported                                             |
| Utility **Compose Text**                                              | ❌ Local graph operation                                 |
| User Input, Batch, Collection, Choose Output, User Result, Annotation | ❌ Sources, control nodes, results, or canvas-only nodes |

The Run action is allowed only when the node's current mode executes on the server and explicitly supports standalone execution. Required model, parameter, and upstream input validation still applies.

## Where the result lives

The result lands **on the node itself**. Each node keeps its own append-only history of its single-node runs, so every time you run it again, the new output is added — it doesn't overwrite the last one. Flip back through previous outputs right there on the node to compare what changed as you tweak settings.

## It is not a workflow run

A single-node run is for iterating, not shipping. It does **not** create a workflow run record.

<Note>
  Nothing from a single-node run appears in [history](/concepts/results-and-history). If you need a run you can find later, share, or reference, do a [full run](/concepts/full-run) instead.
</Note>

|                          | Single-node run           | [Full run](/concepts/full-run)              |
| ------------------------ | ------------------------- | ------------------------------------------- |
| **Runs**                 | One supported server node | Executable nodes in the saved graph         |
| **Result lives**         | On the node               | In [history](/concepts/results-and-history) |
| **Creates a run record** | No                        | Yes                                         |
| **Use it to**            | Iterate on one step       | Produce a final result                      |

Iterate on a node with single runs until it's right, then do a full run to capture the finished output.

## Related

<CardGroup cols={2}>
  <Card title="Running the full workflow" icon="play" href="/concepts/full-run">
    Run the whole graph end to end and produce a run you can keep.
  </Card>

  <Card title="Results & history" icon="clock-rotate-left" href="/concepts/results-and-history">
    Where full runs are recorded, and how to find past outputs.
  </Card>
</CardGroup>