> Source: https://builder.ema.ai/v2/agent-reference/feedback-router
> Title: Feedback Router Agent

# Feedback Router Agent

The Feedback Router agent (`feedback_router`) routes a workflow based on a user's feedback sentiment — **positive** or **negative**. It is a deterministic branch: it reads the sentiment passed into the run, emits it as a routing value, and lets the workflow's edges take it from there. It makes **no LLM call**, so it runs instantly and costs nothing in tokens.

In the AI Employee (AIE) builder, the Feedback Router Agent appears in the **Planning & Resources** group of the agent library.

## When to use it

Use a Feedback Router when a workflow re-runs in response to a user's thumbs-up / thumbs-down feedback and you want to take a different path depending on which it was:

-   On a **negative** branch, regenerate an answer, escalate to a human, or try a different approach.
-   On a **positive** branch, confirm and finish, or record the success.

The Feedback Router only runs on **feedback runs** — runs triggered by a user reacting to a prior result. On a regular (non-feedback) run, the engine skips the Feedback Router node entirely, so it never blocks a normal pass through the workflow.

## How it works

The Feedback Router is resolved locally by the workflow engine — it never dispatches to the agent service, so there is no system prompt, no model, and no version pin.

1.  The engine reads the run's `sentiment` input.
2.  It validates that the value is exactly `"positive"` or `"negative"`. Any other value (including empty) fails the node with a validation error, rather than silently stalling.
3.  It emits the output `{ "feedback": "<sentiment>" }`.
4.  The workflow's outgoing edges match on that `feedback` value to choose the next node.

Because it is purely deterministic, the Feedback Router has no Instructions, no LLM configuration, no knowledge bases, no Tools, and no HITL — it exposes no advanced settings at all.

## Configuration

There is nothing to configure on the node itself. The routing happens on the **edges** leaving the node: you create one edge for the `positive` path and one for the `negative` path, each conditioned on the node's `feedback` output.

In the builder, the Feedback Router node exposes two static route ports — `positive` and `negative` — that you connect to the downstream branches.

## Inputs and output

**Input** — the run supplies `sentiment`, which must be one of two values:

```json
{ "sentiment": "negative" }
```

**Output** — the node echoes the sentiment as `feedback`:

```json
{ "feedback": "negative" }
```

Field

Allowed values

Notes

`sentiment` (input)

`positive`, `negative`

Required. Any other value fails the node with a validation error.

`feedback` (output)

`positive`, `negative`

Echoes the input sentiment. Edge conditions match against this field.

## Example

A support workflow that reacts to user feedback on its answer:

1.  A [Search and Respond](/builder/v2/agent-reference/search-and-respond) node answers the user's question.
2.  The user reacts with a thumbs-up or thumbs-down, triggering a feedback run.
3.  A **Feedback Router** node reads the `sentiment` and emits `feedback`.
4.  The `negative` edge routes to a node that escalates to a human agent; the `positive` edge routes to a confirmation step.

## What's next

-   [Agent Reference overview](/builder/v2/agent-reference) — the type model, catalog, and shared execution engine.
-   [Intent Classification Agent](/builder/v2/agent-reference/intent-classification) — the LLM-based alternative when you need to route on something richer than a thumbs reaction.
-   [Knowledge Search Agent](/builder/v2/agent-reference/knowledge-search) — retrieve and answer from your knowledge base.