dotprompt

Documentation

Overview

Package dotprompt parses and renders dotprompt files.

Index

• func SetDirectory(directory string)
• type Config
• type Prompt
  • func Define(name, templateText string, cfg Config) (*Prompt, error)
  • func New(name, templateText string, cfg Config) (*Prompt, error)
  • func Open(name string) (*Prompt, error)
  • func OpenVariant(name, variant string) (*Prompt, error)
  • func Parse(name, variant string, data []byte) (*Prompt, error)
  • func (p *Prompt) Generate(ctx context.Context, pr *PromptRequest, ...) (*ai.GenerateResponse, error)
  • func (p *Prompt) Register() error
  • func (p *Prompt) RenderMessages(variables map[string]any) ([]*ai.Message, error)
  • func (p *Prompt) RenderText(variables map[string]any) (string, error)
• type PromptRequest

Functions

func SetDirectory

func SetDirectory(directory string)

SetDirectory sets the directory where dotprompt files are read from.

Types

type Config

type Config struct {
	// The prompt variant.
	Variant string
	// The name of the model for which the prompt is input.
	// If this is non-empty, ModelAction should be nil.
	Model string

	// The ModelAction to use.
	// If this is non-nil, Model should be the empty string.
	ModelAction *ai.ModelAction

	// TODO(iant): document
	Tools []*ai.ToolDefinition

	// Number of candidates to generate when passing the prompt
	// to a model. If 0, uses 1.
	Candidates int

	// Details for the model.
	GenerationConfig *ai.GenerationCommonConfig

	InputSchema      *jsonschema.Schema // schema for input variables
	VariableDefaults map[string]any     // default input variable values

	// Desired output format.
	OutputFormat ai.OutputFormat

	// Desired output schema, for JSON output.
	OutputSchema map[string]any // TODO: use *jsonschema.Schema

	// Arbitrary metadata.
	Metadata map[string]any
}

Config is optional configuration for a Prompt.

type Prompt

type Prompt struct {
	// The name of the prompt. Optional unless the prompt is
	// registered as an action.
	Name string

	Config

	// The template for the prompt.
	Template *raymond.Template
	// contains filtered or unexported fields
}

Prompt is a parsed dotprompt file.

A dotprompt file consists of YAML frontmatter within --- lines, followed by a template written in the Handlebars language.

The YAML frontmatter will normally define a JSON schema describing the expected input and output variables. The input variables will appear in the template. The JSON schemas may be defined in a compact picoschema format.

The templates are evaluated with a couple of helpers.

• {{role r}} changes to a new role for the following text
• {{media url=URL}} adds a URL with an optional contentType

func Define

func Define(name, templateText string, cfg Config) (*Prompt, error)

Define creates and registers a new Prompt. This can be called from code that doesn't have a prompt file.

func New

func New(name, templateText string, cfg Config) (*Prompt, error)

New creates a new Prompt without registering it. This may be used for testing or for direct calls not using the genkit action and flow mechanisms.

func Open

func Open(name string) (*Prompt, error)

Open opens and parses a dotprompt file. The name is a base file name, without the ".prompt" extension.

func OpenVariant

func OpenVariant(name, variant string) (*Prompt, error)

OpenVariant opens a parses a dotprompt file with a variant. If the variant does not exist, the non-variant version is tried.

func Parse

func Parse(name, variant string, data []byte) (*Prompt, error)

Parse parses the contents of a dotprompt file.

func (*Prompt) Generate

func (p *Prompt) Generate(ctx context.Context, pr *PromptRequest, cb func(context.Context, *ai.GenerateResponseChunk) error) (*ai.GenerateResponse, error)

Generate executes a prompt. It does variable substitution and passes the rendered template to the AI model specified by the prompt.

This implements the ai.Prompt interface.

func (*Prompt) Register

func (p *Prompt) Register() error

Register registers an action to render a prompt.

func (*Prompt) RenderMessages

func (p *Prompt) RenderMessages(variables map[string]any) ([]*ai.Message, error)

RenderMessages executes the prompt's template and converts it into messages. This just runs the template; it does not call a model.

func (*Prompt) RenderText

func (p *Prompt) RenderText(variables map[string]any) (string, error)

RenderText executes the prompt's template and returns the result as a string. The result may contain only a single, text, message. This just runs the template; it does not call a model.

type PromptRequest

type PromptRequest struct {
	// Input fields for the prompt. If not nil this should be a struct
	// or pointer to a struct that matches the prompt's input schema.
	Variables any `json:"variables,omitempty"`
	// Number of candidates to return; if 0, will be taken
	// from the prompt config; if still 0, will use 1.
	Candidates int `json:"candidates,omitempty"`
	// Model configuration. If nil will be taken from the prompt config.
	Config *ai.GenerationCommonConfig `json:"config,omitempty"`
	// Context to pass to model, if any.
	Context []any `json:"context,omitempty"`
	// The model to use. This overrides any model specified by the prompt.
	Model string `json:"model,omitempty"`
}

PromptRequest is a request to execute a dotprompt template and pass the result to a [ModelAction].