Class: Narrator
Narrator generates meta-content narrations based on other content. It shines at reading and understanding your existing content like articles, help pages, blog posts, etc, and generating short, friendly summaries that tell the reader what content may be most useful to them.
All of the configurations for Narrator are optional, but if you want to generate and save content you
can pass in outputDir and (optionally) outputFilename to have Narrator automatically save its
generations for you. For example, if we want to save our generated content as .md files in
the ./editorial directory, we can configure it like this:
export const narrator = new Narrator({
outputFilename: (docId) => `${docId}.md`,
outputDir: path.join(process.cwd(), "editorial"),
});
Now we can generate some content, which in this case will be saved to ./editorial/tag/ai.md
(directories will be created for you):
const content = await narrator.generate(
{
docId: "tag/ai",
suffix: "Please reply with only the markdown you generate", //suffix is optional
prompt: `
These are summaries of my most recent articles about AI. Your task is to generate a 2-3 sentence
introduction that tells readers at-a-glance what I've been writing about. Please generate markdown,
and include links to the articles. Do not use triple backticks or section headings in your response.
<<Articles go here>>
`,
},
{ save: true }
);
This will generate content for you and save it according to the configuration you provided. You can set
docId to whatever you want - in this case we're generating intro text for a blog that contains
articles about AI. If you don't specify a model it will default
to using OpenAI's "gpt-4o", but you can pass in any model provided by the
Vercel AI SDK.
Constructors
new Narrator()
new Narrator(
args):Narrator
Creates a new instance of the Narrator class.
Parameters
• args: NarratorArgs
The arguments to configure the Narrator instance.
Returns
Defined in
Properties
examplesDir
examplesDir:
undefined|string
Directory where example YAML files are stored. Used for fetching good and bad examples to guide model generation.
Defined in
exampleTemplate
exampleTemplate:
Function
Function to format examples.
This function is applied to examples before they are used in prompts.
Example function signature: (example: Example) => string.
Defined in
logger
logger:
Logger
Custom logger instance for logging (e.g., a Winston logger). If undefined, a default logger will be used.
Defined in
model
model:
any
The model used for text generation, such as an instance of OpenAI's GPT models.
If undefined, a default model (e.g., openai("gpt-4o")) will be used.
Defined in
outputDir
outputDir:
undefined|string
Directory where generated outputs will be saved. If undefined, outputs will not be saved automatically.
Defined in
outputFilename
outputFilename:
Function
Function to generate output filenames based on the document ID.
If undefined, filenames are generated using the document ID by default.
Example function signature: (docId: string) => string.
Defined in
parallel
parallel:
number=1
Number of parallel processes to use when running generateMulti.
Defaults to 1 if not provided.
Defined in
temperature
temperature:
any
Temperature for the model generation. Controls the randomness of the generated output, where a higher value results in more varied responses. If undefined, the default temperature is used.
Defined in
trainer
trainer:
Trainer
Trainer instance responsible for evaluating generated examples. Used to classify examples as good or bad based on feedback or custom logic.
Defined in
Methods
generate()
generate(
task,options):Promise<string|AsyncIterableStream<string>>
Generates content for the provided task based on the given options.
Parameters
• task: GenerationTask
The task containing the prompt and document ID.
• options: GenerateOptions = {}
Optional settings for generation (temperature, model, etc.).
Returns
Promise<string | AsyncIterableStream<string>>
The generated text content.
Defined in
getExampleGroup()
getExampleGroup(
docId?):undefined|string
Extracts the first part of a document ID, representing the example group.
Parameters
• docId?: string
The document ID.
Returns
undefined | string
The group key.
Defined in
getExampleKey()
getExampleKey(
docId):string
Extracts the key from a document ID.
Parameters
• docId: string
The document ID.
Returns
string
The extracted key.
Defined in
getExamplesForTask()
getExamplesForTask(
task,verdict,limit?):Promise<Example[]>
Retrieves examples for a specific task based on the verdict (good or bad).
Parameters
• task: GenerationTask
The generation task for which examples are retrieved.
• verdict: string
The type of examples to retrieve ("good" or "bad").
• limit?: number
The maximum number of examples to retrieve (optional).
Returns
Promise<Example[]>
An array of examples.
Defined in
getNarration()
getNarration(
docId):string|false
Retrieves previously saved narration by document ID.
Parameters
• docId: string
The document ID of the narration.
Returns
string | false
The narration content or false if the content is not found or could not be read.
Defined in
readExampleYaml()
readExampleYaml(
filename):any
Reads a YAML file and parses its contents.
Parameters
• filename: string
The path to the YAML file.
Returns
any
The parsed YAML content or an empty array in case of an error.
Defined in
saveExample()
saveExample(
args):Promise<boolean>
Saves the evaluation result as an example for future reference.
Parameters
• args: SaveExampleArgs
The arguments for saving the example (document ID, content, etc.).
Returns
Promise<boolean>
A boolean indicating success or failure of the save operation.
Defined in
saveNarration()
saveNarration(
narration):boolean
Saves the generated narration to the output directory.
Parameters
• narration: Narration
The narration object containing document ID and content.
Returns
boolean
A boolean indicating success or failure of the save operation.
Defined in
train()
train(
task):Promise<undefined|boolean>
Trains the Narrator using a task and evaluates the result. If the result is marked for saving, it will be saved.
Parameters
• task: GenerationTask
The generation task to be processed.
Returns
Promise<undefined | boolean>