Documentation
¶
Overview ¶
Package predict provides LLM-based prediction functionality for the gsh REPL. It includes prefix-based prediction, null-state prediction, command explanation, and a router to coordinate between different prediction strategies.
Index ¶
Constants ¶
const BestPractices = `* Git commit messages should follow conventional commit message format`
BestPractices contains shell command best practices used in prediction prompts.
Variables ¶
This section is empty.
Functions ¶
This section is empty.
Types ¶
type ContextFormatter ¶
type ContextFormatter interface {
// FormatContext formats the context map into a string.
FormatContext(contextMap map[string]string) string
}
ContextFormatter formats a context map into a string suitable for LLM prompts.
type DefaultContextFormatter ¶
type DefaultContextFormatter struct{}
DefaultContextFormatter is the default implementation of ContextFormatter.
func (*DefaultContextFormatter) FormatContext ¶
func (f *DefaultContextFormatter) FormatContext(contextMap map[string]string) string
FormatContext formats the context map into a string with labeled sections.
type NullStatePredictor ¶
type NullStatePredictor struct {
// contains filtered or unexported fields
}
NullStatePredictor predicts commands when the input is empty. It uses context (cwd, git status, history, etc.) to suggest a likely next command.
func NewNullStatePredictor ¶
func NewNullStatePredictor(cfg NullStatePredictorConfig) *NullStatePredictor
NewNullStatePredictor creates a new NullStatePredictor with the given configuration.
func (*NullStatePredictor) Predict ¶
Predict returns a prediction for the next command when input is empty.
func (*NullStatePredictor) UpdateContext ¶
func (p *NullStatePredictor) UpdateContext(contextMap map[string]string)
UpdateContext updates the context information used for predictions.
type NullStatePredictorConfig ¶
type NullStatePredictorConfig struct {
// ModelResolver resolves to an LLM model for predictions.
// Can be a direct ModelValue or an SDKModelRef for lazy resolution.
ModelResolver interpreter.ModelResolver
// Logger for debug output. If nil, a no-op logger is used.
Logger *zap.Logger
// Formatter for context text. If nil, DefaultContextFormatter is used.
Formatter ContextFormatter
}
NullStatePredictorConfig holds configuration for creating a NullStatePredictor.
type Predictor ¶
type Predictor interface {
// Predict returns a prediction for the given input.
// The context can be used for cancellation.
// Returns the predicted command and any error that occurred.
Predict(ctx context.Context, input string) (prediction string, err error)
// UpdateContext updates the context information used for predictions.
// The context map contains key-value pairs like "cwd", "git", "history", etc.
UpdateContext(contextMap map[string]string)
}
Predictor defines the interface for making command predictions. Implementations can use different strategies (history, LLM, etc.)
type PrefixPredictor ¶
type PrefixPredictor struct {
// contains filtered or unexported fields
}
PrefixPredictor predicts command completions based on a partial command prefix. It uses an LLM to generate predictions that start with the given prefix.
func NewPrefixPredictor ¶
func NewPrefixPredictor(cfg PrefixPredictorConfig) *PrefixPredictor
NewPrefixPredictor creates a new PrefixPredictor with the given configuration.
func (*PrefixPredictor) Predict ¶
Predict returns a prediction for the given input prefix. The prediction will start with the input prefix.
func (*PrefixPredictor) UpdateContext ¶
func (p *PrefixPredictor) UpdateContext(contextMap map[string]string)
UpdateContext updates the context information used for predictions.
type PrefixPredictorConfig ¶
type PrefixPredictorConfig struct {
// ModelResolver resolves to an LLM model for predictions.
// Can be a direct ModelValue or an SDKModelRef for lazy resolution.
ModelResolver interpreter.ModelResolver
// Logger for debug output. If nil, a no-op logger is used.
Logger *zap.Logger
// Formatter for context text. If nil, DefaultContextFormatter is used.
Formatter ContextFormatter
}
PrefixPredictorConfig holds configuration for creating a PrefixPredictor.
type Router ¶
type Router struct {
// contains filtered or unexported fields
}
Router coordinates between different prediction strategies. It routes prediction requests to the appropriate predictor based on the input state.
func NewRouter ¶
func NewRouter(cfg RouterConfig) *Router
NewRouter creates a new Router with the given configuration.
func NewRouterFromConfig ¶
func NewRouterFromConfig(modelResolver interpreter.ModelResolver, logger *zap.Logger) *Router
NewRouterFromConfig creates a Router configured from a model resolver. It sets up the prefix and null-state predictors using the provided model resolver. Returns nil if no model resolver is provided.
func (*Router) NullStatePredictor ¶
NullStatePredictor returns the null state predictor (for testing).
func (*Router) Predict ¶
Predict routes the prediction request to the appropriate predictor. If input is empty, uses null state predictor; otherwise uses prefix predictor.
func (*Router) PrefixPredictor ¶
PrefixPredictor returns the prefix predictor (for testing).
func (*Router) UpdateContext ¶
UpdateContext updates the context for all predictors.
type RouterConfig ¶
type RouterConfig struct {
// PrefixPredictor handles predictions when there is input text.
PrefixPredictor Predictor
// NullStatePredictor handles predictions when input is empty.
NullStatePredictor Predictor
// Logger for debug output. If nil, a no-op logger is used.
Logger *zap.Logger
}
RouterConfig holds configuration for creating a Router.
Source Files