Skip to content

integrations.md

Latest commit

 

History

History
91 lines (63 loc) · 2.65 KB

integrations.md

File metadata and controls

91 lines (63 loc) · 2.65 KB
Raw

Tracing integrations

Auto-instrument popular frameworks so their calls become PromptRails spans. Each integration only needs its own optional dependency.

pip install "promptrails[langchain]"   # or [openai], [anthropic], [google], [otel]

LangChain

A callback handler that turns LangChain runs (chains, LLMs, tools, retrievers) into a span tree. Pass it via callbacks:

from promptrails.tracing import Tracer
from promptrails.tracing.integrations.langchain import PromptRailsCallbackHandler

tracer = Tracer(api_key="pr_...")
handler = PromptRailsCallbackHandler(tracer)

chain.invoke({"question": "What's the weather?"}, config={"callbacks": [handler]})
tracer.flush()

The tree is built from LangChain's run_id/parent_run_id, so it is correct under threads and async runs. Token usage and model are read from on_llm_end.

OpenAI

Wrap an OpenAI-compatible client so every chat.completions.create call emits an llm span with model, token usage, latency, and output:

from openai import OpenAI
from promptrails.tracing import Tracer
from promptrails.tracing.integrations.openai import trace_openai

tracer = Tracer(api_key="pr_...")
client = trace_openai(OpenAI(), tracer)

client.chat.completions.create(model="gpt-4o", messages=[{"role": "user", "content": "hi"}])

The wrapper is duck-typed, so it also works with any API-compatible client.

Anthropic

Wrap an Anthropic client to trace every messages.create call:

from anthropic import Anthropic
from promptrails.tracing.integrations.anthropic import trace_anthropic

client = trace_anthropic(Anthropic(), tracer)
client.messages.create(model="claude-sonnet-4-5", max_tokens=1024, messages=[...])

Install with pip install "promptrails[anthropic]".

Google GenAI

Wrap a Google GenAI client (the unified google-genai SDK) to trace every models.generate_content call:

from google import genai
from promptrails.tracing.integrations.google import trace_google

client = trace_google(genai.Client(), tracer)
client.models.generate_content(model="gemini-2.0-flash", contents="Hello")

Install with pip install "promptrails[google]".

OpenTelemetry

Already using OpenTelemetry? Register the PromptRails exporter and your existing spans flow in — gen_ai.* semantic-convention attributes are mapped onto the span model:

from opentelemetry.sdk.trace import TracerProvider
from opentelemetry.sdk.trace.export import BatchSpanProcessor
from promptrails.tracing.integrations.otel import PromptRailsSpanExporter

provider = TracerProvider()
provider.add_span_processor(
    BatchSpanProcessor(PromptRailsSpanExporter(api_key="pr_..."))
)