LLMMetaData

@Serializable
data class LLMMetaData(val port: Int = 11434, val model: String = "qwen2.5-coder:32b-instruct-q8_0", val prompt: String = "", val backend: LlmBackend = LlmBackend.OLLAMA, val systemPrompt: String = "", val responseFormat: ResponseFormat = ResponseFormat.NATURAL_LANGUAGE, val responseInstructions: String = LLMResult.JSON_SCHEMA, val error: String = "", val sources: List<NodeIdentity> = emptyList(), val snapshot: Snapshot = Snapshot(), val invocationTriggers: List<InvocationTrigger> = emptyList(), val nodeAction: NodeAction = NodeAction.EXECUTE, val inputs: List<NodeIdentity> = emptyList()) : SourceMetaData(source)

Payload for a Server.LLM node.

Constructors

Link copied to clipboard
constructor(port: Int = 11434, model: String = "qwen2.5-coder:32b-instruct-q8_0", prompt: String = "", backend: LlmBackend = LlmBackend.OLLAMA, systemPrompt: String = "", responseFormat: ResponseFormat = ResponseFormat.NATURAL_LANGUAGE, responseInstructions: String = LLMResult.JSON_SCHEMA, error: String = "", sources: List<NodeIdentity> = emptyList(), snapshot: Snapshot = Snapshot(), invocationTriggers: List<InvocationTrigger> = emptyList(), nodeAction: NodeAction = NodeAction.EXECUTE, inputs: List<NodeIdentity> = emptyList())

Properties

Link copied to clipboard

Inference backend this node routes requests to.

Link copied to clipboard
open override val error: String

Last known error message for this node, or empty string when healthy.

Link copied to clipboard
open override val inputs: List<NodeIdentity>

sets the nodes this node reads from to complete its work when invoked

Link copied to clipboard

The set of InvocationTriggers configured to wake this node. The node processor checks the incoming event against this list before doing any work. An empty list means "never auto-fire" — only manual execution.

Link copied to clipboard

Model identifier sent on every request (e.g. "qwen2.5-coder:32b-instruct-q8_0").

Link copied to clipboard
open override val nodeAction: NodeAction

The action this node performs when it fires. Defaults to NodeAction.EXECUTE on every concrete implementation so that deserialising a payload that predates this field yields the original behaviour unchanged.

Link copied to clipboard
val port: Int

Port on the server hosting the inference endpoint.

Link copied to clipboard

User-entered prompt template; injected into the request at invocation time.

Link copied to clipboard

How the model should format its reply.

Link copied to clipboard

JSON Schema (or natural-language instruction) the model must follow when responseFormat is ResponseFormat.JSON. Defaults to LLMResult.JSON_SCHEMA so observer nodes can decode snapshot.value as LLMResult out of the box.

Link copied to clipboard
open override val snapshot: Snapshot

Stores the last result of when this node was invoked and is a source of data

Link copied to clipboard
open override val sources: List<NodeIdentity>

Upstream nodes whose values feed this one. For a filter or executor this is the data being read; for a trigger it is the value being watched.

Link copied to clipboard

System prompt prepended to every request. Blank means the server applies its default Krill persona automatically.