Quickstart

Overview

respan is the recommended way to use Respan. It provides a single Respan() class that initializes tracing, captures LLM calls, and exposes decorators for structured spans.

$
pip install respan-ai

Version: 3.0.0 | Python: >=3.9, <4.0

Quick start

With auto_instrument=True, Respan auto-traces any installed LLM SDK (OpenAI, Anthropic, LangChain, vector DBs, and more) — no instrumentor imports required.

1
from respan import Respan
2
3
respan = Respan(api_key="your-api-key", auto_instrument=True)
4
5
from openai import OpenAI
6
OpenAI().chat.completions.create(
7
    model="gpt-4o",
8
    messages=[{"role": "user", "content": "Hello"}],
9
)

That’s it — the call above appears in your Respan dashboard.

Explicit instrumentation (optional)

Prefer to pin a specific instrumentation plugin (for finer control or to avoid auto-discovering a library you don’t want traced)? Pass it via instrumentations:

1
from respan import Respan
2
from respan_instrumentation_openai_agents import OpenAIAgentsInstrumentor
3
4
respan = Respan(
5
    api_key="your-api-key",
6
    instrumentations=[OpenAIAgentsInstrumentor()],
7
)

You can combine auto_instrument=True with explicit plugins — plugins take precedence.

See Respan() for every constructor parameter, environment variables, and more examples.

Public exports

Everything you need is available from respan:

1
from respan import (
2
    # Core
3
    Respan,
4
    Instrumentation,
5
6
    # Decorators
7
    workflow, task, agent, tool,
8
9
    # Client
10
    RespanClient, get_client,
11
12
    # Context managers
13
    propagate_attributes,
14
    respan_span_attributes,
15
)

Next steps

• Respan() — constructor parameters and configuration examples
• propagate_attributes() — attach user/thread/metadata to all spans
• log_batch_results() — log OpenAI Batch API results
• Instrumentation Protocol — build your own plugin