All nodes/AI/RAG

Reranker

Reranker configuration (Cohere/Jina/Voyage). Connects by link to the Rerank node. Performs no computation itself.

Try it

When to use it

This node ranks nothing — it names the model and the Top N for Rerank. It has no other consumer: the link_rerank wire is accepted by that node alone.

If there is a single rerank in the graph, set the model and Top N right in its form — a separate node is not needed then. It pays off where several reranks exist (one per branch of a hybrid search, say) and their settings must stay in sync.

How it works

There are no inputs and no run port: the node never becomes a step of its own, and its value is computed when a consumer asks for it. Three things travel through the port: the type, the model id and Top N.

Both values override the same-named fields on the Rerank node, which go inactive in its form. Top N has a subtlety: it always travels through the port — even if you never touched it here, the default goes out rather than whatever the consumer had. The model is gentler: an empty value here does not blank the setting, the consumer falls back to its own field.

The model list is the platform catalog plus the models of your connections whose purpose is Rerank. Tokens spent on your own key are not billed by the platform.

Common mistakes

• Expecting the node to run. It is configuration, not an action: no step of its own appears in the run log.
• Editing Top N on the Rerank node while a Reranker is connected. The value arrives from here, and the edit on the consumer has no effect.
• One node for two different needs. If one branch needs 3 results and another needs 10, use two Reranker nodes: the settings travel as a set.
• Looking for the model in the LLM list. Rerankers are a separate model catalog and have nothing to do with choosing an LLM.

Inputs

—

Outputs

Configuration