Framework Integrations

Strathon instruments your agent framework's own extension points — callback handlers, plugins, hooks, event listeners. Most integrations connect with a single instrument() call; the callback-, plugin-, hook-, and capability-based frameworks (LangGraph, LangChain, Google ADK, Claude Agent SDK, Pydantic AI) take one extra line to attach the handler, plugin, hooks, or capability — each guide shows the exact pattern. None of it changes your agent logic.

bash

pip install "strathon[langgraph]"   # one framework
pip install "strathon[all]"         # all 10

Then in your code:

python

from strathon import Client
from strathon.instrumentation.langgraph import instrument

client = Client(api_key="stra_...", endpoint="http://localhost:4318")
handler = instrument(client)

# LangGraph returns a callback handler — attach it on each invocation:
result = agent.invoke(inputs, config={"callbacks": [handler]})

That is the LangGraph pattern; CrewAI, the OpenAI Agents SDK, and AutoGen need only instrument(client, frameworks=["..."]), while Google ADK, the Claude Agent SDK, and Pydantic AI each take a one-time wiring step shown in their guide.

Supported frameworks

Framework	Integration	Guide
LangGraph	LangChain `BaseCallbackHandler`: intercepts tool calls before execution	Guide
CrewAI	Event listener (tracing) + tool-invoke wrapping (enforcement)	Guide
LangChain	Same callback handler as LangGraph	Guide
OpenAI Agents SDK	`TracingProcessor` (tracing) + `RunHooks` (enforcement)	Guide
Google ADK	First-class `BasePlugin`	Guide
Pydantic AI	First-class `AbstractCapability`	Guide
Claude Agent SDK	`PreToolUse`/`PostToolUse` hooks (tracing via `query()` wrapper)	Guide
AutoGen	`BaseTool.run_json` wrapper (tool enforcement) + conversation tracing	Guide
OpenAI	Drop-in wrapper for `chat.completions.create`	Guide
Anthropic	Drop-in wrapper for `messages.create`	Guide

Which should I use?

If you are building production agents and not already committed to a framework, LangGraph is the most widely deployed choice for stateful, auditable workflows and is the best-supported integration here. If you already use a framework, pick its guide above; every integration captures the same trace data and enforces the same policies.

The raw model-SDK wrappers (OpenAI, Anthropic) instrument direct model calls rather than an agent framework. Use them when you call a model SDK directly without an orchestration layer.