LLM Enrichment

The LLM node sends asset metadata to a Large Language Model and stores the model’s response as enrichment — for example a classification, a set of suggested tags, or a description. Multiple prompts can be configured, each producing its own output.

Kind

llm

Applies to

Any (operates on the asset’s filename / metadata)

Inputs

Optional upstream outputs (e.g. a transcript or extracted text) can be referenced by prompts

Output keys

llm_result_{promptId} (String) — one per configured prompt

Requirements

An LLM provider service reachable from the worker. The node supports an OpenAI-compatible (vLLM) endpoint; CPU on the worker, the model runs in the service.

Persists to

asset_json_comp per prompt + asset_node_result ledger

Configuration

LLMNodeOptions:

Option Meaning

Option	Meaning
provider URL	Base URL of the LLM provider service
`providerType`	Provider protocol — the default local provider, or `VLLM` for an OpenAI-compatible endpoint
`prompts`	A map of prompt definitions; each produces a `llm_result_{promptId}` output

provider URL

Base URL of the LLM provider service

providerType

Provider protocol — the default local provider, or VLLM for an OpenAI-compatible endpoint

prompts

A map of prompt definitions; each produces a llm_result_{promptId} output

Use Cases

Auto-classification — categorise assets from their metadata or transcript.
Tag suggestions — propose tags for review.
Description generation — draft human-readable summaries stored alongside the asset.