Classification Synapse

Categorize text into predefined classes.

Constructor

func Classification(task string, categories []string, provider Provider, opts ...Option) (*ClassificationSynapse, error)

Parameters:

• task - Description of what to classify
• categories - Valid category options
• provider - LLM provider
• opts - Optional configuration

Returns:

• *ClassificationSynapse - The configured synapse
• error - Configuration error

Methods

Fire

func (s *ClassificationSynapse) Fire(ctx context.Context, session *Session, input string) (string, error)

Execute and return the primary category.

Returns:

• string - Primary category (one of the provided categories)
• error - Execution error

FireWithInput

func (s *ClassificationSynapse) FireWithInput(ctx context.Context, session *Session, input ClassificationInput) (string, error)

Execute with structured input including examples.

FireWithDetails

func (s *ClassificationSynapse) FireWithDetails(ctx context.Context, session *Session, input string) (*ClassificationResponse, error)

Execute and return full response.

Types

type ClassificationInput struct {
    Subject  string              // Text to classify
    Examples map[string][]string // Optional examples per category
}

type ClassificationResponse struct {
    Primary    string   `json:"primary"`
    Secondary  string   `json:"secondary,omitempty"`
    Confidence float64  `json:"confidence"`
    Reasoning  []string `json:"reasoning"`
}

Examples

Basic Usage

classifier, _ := zyn.Classification(
    "What type of email is this?",
    []string{"spam", "urgent", "newsletter", "personal"},
    provider,
)
session := zyn.NewSession()

result, err := classifier.Fire(ctx, session, "URGENT: Your account will be suspended!")
// result: "urgent"

With Examples

input := zyn.ClassificationInput{
    Subject: "Great product, fast shipping!",
    Examples: map[string][]string{
        "positive": {"Love it!", "Excellent quality"},
        "negative": {"Terrible", "Waste of money"},
        "neutral":  {"It's okay", "As expected"},
    },
}

result, err := classifier.FireWithInput(ctx, session, input)
// result: "positive"

With Details

response, err := classifier.FireWithDetails(ctx, session, "Check out this amazing deal!")
// response.Primary: "spam"
// response.Secondary: "promotional"
// response.Confidence: 0.87
// response.Reasoning: ["Contains promotional language", "Urgency tactics"]

Use Cases

• Email routing
• Support ticket triage
• Content categorization
• Sentiment analysis (simple)
• Intent detection