Integrations

Splyntra ingests agent telemetry over OpenTelemetry (OTLP/HTTP) and understands the GenAI semantic conventions. That means anything that can emit OTel can send data to Splyntra, and the popular agent frameworks work with a one-line SDK init.

All examples assume you've set:

export SPLYNTRA_ENDPOINT="https://ingest.splyntra.com"   # or http://localhost:4318 self-hosted
export SPLYNTRA_API_KEY="sk_…"

Framework	Method	Notes
LangGraph / LangChain	splyntra.instrument()	Nodes, tools, and model calls traced
CrewAI	splyntra.instrument()	Agents and tasks become spans
OpenAI Agents SDK	splyntra.instrument()	Runners, handoffs, tools
LlamaIndex	splyntra.instrument()	Query engines and retrievers
AutoGen	splyntra.instrument()	Conversable agents and group chats
Dify	OTLP exporter	Configure the built-in OTel exporter
n8n	OTLP / HTTP node	Emit spans from workflow nodes
Any OTel source	OTLP/HTTP	Point your exporter at the endpoint

LangGraph

from splyntra import instrument
instrument(endpoint=..., api_key=..., project="support-agent")

from langgraph.prebuilt import create_react_agent
agent = create_react_agent(model, tools)
agent.invoke({"messages": [("user", "Where is my order?")]})

Each graph node, tool call, and model invocation is captured as a span in the run tree.

CrewAI

from splyntra import instrument
instrument(endpoint=..., api_key=..., project="ops-crew")

crew.kickoff()   # each agent and task is traced automatically

OpenAI Agents SDK

from splyntra import instrument
instrument(endpoint=..., api_key=..., project="research-agent")

from agents import Agent, Runner
Runner.run_sync(Agent(name="researcher", instructions="Cite sources."), "…")

Handoffs between agents show up as nested spans, so multi-agent flows stay readable.

LlamaIndex

from splyntra import instrument
instrument(endpoint=..., api_key=..., project="rag-app")

# Query engines, retrievers, and LLM calls are traced.
response = query_engine.query("What changed in the Q3 policy?")

AutoGen

from splyntra import instrument
instrument(endpoint=..., api_key=..., project="autogen-lab")

# Conversable agents and group chats are captured as spans.

Dify

Dify can export OpenTelemetry directly. In your Dify deployment, enable the OTLP exporter and set the endpoint and headers:

OTEL_EXPORTER_OTLP_ENDPOINT=https://ingest.splyntra.com
OTEL_EXPORTER_OTLP_HEADERS=x-splyntra-api-key=${SPLYNTRA_API_KEY}

n8n

Use an HTTP Request node (or the OpenTelemetry community node) to emit spans for each workflow step to the OTLP/HTTP endpoint, passing the API key as the x-splyntra-api-key header.

Raw OpenTelemetry

Any language or framework with an OTel SDK can send data — no Splyntra SDK required:

export OTEL_EXPORTER_OTLP_ENDPOINT="https://ingest.splyntra.com"
export OTEL_EXPORTER_OTLP_HEADERS="x-splyntra-api-key=${SPLYNTRA_API_KEY}"
export OTEL_SERVICE_NAME="my-agent"

tip

Emit the standard gen_ai.* attributes (model, token usage, prompts) and Splyntra will populate cost, model breakdowns, and detection with no extra configuration.

Next steps

• Quickstart — end-to-end setup.
• Traces & risk — what the ingested spans become.