Class VercelModelRunner

Runner implementation for Vercel AI SDK chat models.

Implements the unified Runner protocol via run. Returned by VercelRunnerFactory.createModel.

Hierarchy

VercelModelRunner

Implements

Runner

Index

Constructors

Properties

_history _logger? _model _multiTurn _parameters

Methods

_runCompletion _runStructured getModel run

Constructors

constructor

new VercelModelRunner(model, config, parameters, logger?, multiTurn?): VercelModelRunner
Parameters
- model: LanguageModel
- config: LDAICompletionConfig
- parameters: VercelAIModelParameters
- Optional logger: LDLogger
  Optional
- multiTurn: boolean = true
Returns VercelModelRunner
- Defined in ai-providers/server-ai-vercel/src/VercelModelRunner.ts:26

Properties

`Private` _history

_history: ModelMessage[]

`Private` `Optional` _logger

_logger?: LDLogger

`Private` _model

_model: LanguageModel

`Private` _multiTurn

_multiTurn: boolean

`Private` _parameters

_parameters: VercelAIModelParameters

Methods

`Private` _runCompletion

_runCompletion(messages): Promise<RunnerResult>
Parameters
- messages: ModelMessage[]
Returns Promise<RunnerResult>
- Defined in ai-providers/server-ai-vercel/src/VercelModelRunner.ts:83

`Private` _runStructured

_runStructured(messages, outputType): Promise<RunnerResult>
Parameters
- messages: ModelMessage[]
- outputType: Record<string, unknown>
Returns Promise<RunnerResult>
- Defined in ai-providers/server-ai-vercel/src/VercelModelRunner.ts:103

getModel

getModel(): LanguageModel
Get the underlying Vercel AI model instance.

Returns LanguageModel
- Defined in ai-providers/server-ai-vercel/src/VercelModelRunner.ts:79

run

run(input, outputType?): Promise<RunnerResult>
Run the Vercel AI model with the given user prompt.

The runner maintains a conversation history (as Vercel AI SDK ModelMessages) that is initialized from any messages on the AI config (system prompt, etc.). On every invocation the user prompt is appended to the existing history before being sent to the model. When multiTurn is true (the default) and the call succeeds with non-empty content, the user prompt and the assistant's reply are persisted to the history so subsequent calls continue the conversation. When multiTurn is false, history is treated as read-only — useful for stateless runners (e.g. judges) where every call should see only the initial config messages plus the current input. Failed calls leave the history unchanged so the next call can retry cleanly.
Parameters
- input: string
  
  The user prompt string.
- Optional outputType: Record<string, unknown>
  
  Optional JSON schema for structured output. When provided, the parsed object is exposed via RunnerResult.parsed.
  
  Optional
Returns Promise<RunnerResult>
Implementation of Runner.run
- Defined in ai-providers/server-ai-vercel/src/VercelModelRunner.ts:59

Generated using TypeDoc