Module: Langfuse::Types

Defined in:: lib/langfuse/types.rb

Overview

Type definitions and constants for Langfuse domain models

Provides observation type constants and attribute classes for type-safe handling of Langfuse observations and traces.

Examples:

Using observation types

Langfuse::Types::OBSERVATION_TYPES # => ["span", "generation", ...]
Langfuse::Types::LEVELS # => ["DEBUG", "DEFAULT", "WARNING", "ERROR"]

Using attribute classes

attrs = Langfuse::Types::SpanAttributes.new(
  input: { query: "test" },
  level: "DEFAULT",
  metadata: { source: "api" }
)
attrs.to_h # => { input: {...}, level: "DEFAULT", metadata: {...} }

Defined Under Namespace

Classes: EmbeddingAttributes, GenerationAttributes, SpanAttributes, TraceAttributes

Constant Summary collapse

OBSERVATION_TYPES = Types of observations that can be created in Langfuse span: General-purpose observations for tracking operations, functions, or logical units of work generation: Specialized observations for LLM calls with model parameters, usage, and costs event: Point-in-time occurrences or log entries within a trace embedding: Observations for embedding generation calls agent: Observations for agent-based workflows tool: Observations for tool/function calls chain: Observations for chain-based workflows retriever: Observations for retrieval operations evaluator: Observations for evaluation operations guardrail: Observations for guardrail checks Returns: (Array<String>) — Array of observation type strings

%w[
  span generation event embedding
  agent tool chain retriever
  evaluator guardrail
].freeze

LEVELS = Severity levels for observations in Langfuse Used to categorize the importance or severity of observations: DEBUG: Detailed diagnostic information DEFAULT: Normal operation information WARNING: Potentially problematic situations ERROR: Error conditions that need attention Returns: (Array<String>) — Array of level strings

%w[DEBUG DEFAULT WARNING ERROR].freeze

SCORE_DATA_TYPES = Data types for Langfuse scores Used to categorize score values: numeric: Numeric values (Integer, Float, BigDecimal) boolean: Boolean values (true/false or 0/1) categorical: String values for categorical scores text: Free-form String values (1 to 500 characters) correction: Corrected output String attached to a trace or observation (conventionally named "output") Returns: (Hash<Symbol, String>) — Hash mapping symbol keys to API string values

{
  numeric: "NUMERIC",
  boolean: "BOOLEAN",
  categorical: "CATEGORICAL",
  text: "TEXT",
  correction: "CORRECTION"
}.freeze

VALID_SCORE_DATA_TYPES = Valid score data type symbols Returns: (Array<Symbol>) — Array of valid data type symbols

SCORE_DATA_TYPES.keys.freeze

EXPERIMENT_SCORE_DATA_TYPES = Score data types accepted by experiment evaluations Experiment evaluations represent metrics, so this set is intentionally narrower than SCORE_DATA_TYPES: text notes and corrected outputs are general scores, not experiment metrics (matching langfuse-python). Returns: (Array<Symbol>) — Array of experiment-safe data type symbols

%i[numeric boolean categorical].freeze

EventAttributes = Alias for event observation attributes (same as SpanAttributes)

SpanAttributes

AgentAttributes = Alias for agent observation attributes (same as SpanAttributes)

SpanAttributes

ToolAttributes = Alias for tool observation attributes (same as SpanAttributes)

SpanAttributes

ChainAttributes = Alias for chain observation attributes (same as SpanAttributes)

SpanAttributes

RetrieverAttributes = Alias for retriever observation attributes (same as SpanAttributes)

SpanAttributes

EvaluatorAttributes = Alias for evaluator observation attributes (same as SpanAttributes)

SpanAttributes

GuardrailAttributes = Alias for guardrail observation attributes (same as SpanAttributes)

SpanAttributes