Skip to main content
Weave is built on three foundational primitives that work together across all features:

​
Ops

An Op is a versioned, tracked function. When you decorate a function with @weave.op() (Python) or wrap it with weave.op() (TypeScript), Weave automatically captures its code, inputs, outputs, and execution metadata. 
import weave

@weave.op()
def generate_response(prompt: str) -> str:
    # Your LLM logic here
    return response
Ops are the building blocks of tracing, evaluation scorers, and any tracked computation.

​
Objects

An Object is versioned, serializable data. Weave automatically versions objects when they change, creating an immutable history. Objects include:
  • Datasets: Collections of examples for evaluation
  • Models: Configurations and parameters for your LLM logic
  • Prompts: Versioned prompt templates
dataset = weave.Dataset(
    name="test-cases",
    rows=[
        {"input": "What is 2+2?", "expected": "4"},
        {"input": "What is the capital of France?", "expected": "Paris"},
    ]
)
weave.publish(dataset)

​
Calls

A Call is a logged execution of an Op. Every time an Op runs, Weave creates a Call that captures:
  • Input arguments
  • Output value
  • Timing and latency
  • Parent-child relationships (for nested calls)
  • Any errors that occurred
Calls form the backbone of Weave’s tracing system and provide the data for debugging, analysis, and evaluation.