Integrations Overview

​

Integration Matrix

​

Integration Patterns

​

Python

import cascadeflow
cascadeflow.init(mode="observe")

# Framework-specific activation
from cascadeflow.integrations.langchain import get_harness_callback
from cascadeflow.integrations.openai_agents import CascadeFlowModelProvider
from cascadeflow.integrations.crewai import enable as enable_crewai
from cascadeflow.integrations.google_adk import enable as enable_adk
from cascadeflow.integrations.pydantic_ai import create_cascade_model
from cascadeflow.integrations.hermes import HermesDelegationRouter

​

TypeScript

npm install @cascadeflow/langchain
npm install @cascadeflow/vercel-ai
npm install @cascadeflow/n8n-nodes-cascadeflow

​

Choosing an Integration

• LangChain/LangGraph: Use if you have existing LangChain chains or agents. The callback handler wraps any BaseChatModel.
• OpenAI Agents SDK: Use if you’re building with OpenAI’s Agents SDK. The ModelProvider supports model candidates and tool gating.
• CrewAI: Use if you’re building multi-agent crews. The llm_hooks integration tracks all crew steps.
• Google ADK: Use if you’re building with Google’s Agent Development Kit. The plugin integrates with Runner.
• PydanticAI: Use if you’re building with PydanticAI agents. The cascade Model performs drafter→verifier routing with quality gating and tool risk.
• n8n: Use if you’re building no-code workflows. The community node adds cascade routing to any n8n flow.
• Vercel AI SDK: Use if you’re building TypeScript server-side agents. The middleware wraps AI SDK streams.
• Hermes Agent: Use if you need per-skill, complexity-aware, or topic-aware routing for delegated subagents while Hermes keeps provider credentials and fallback behavior.

​

What Stays Consistent Across Frameworks

• The harness sees runtime state inside the workflow, not only the request boundary
• Budgets, traces, and policy logic remain first-class across integrations
• The goal is governable agent behavior, not isolated cost routing
• GitHub examples remain the secondary deep-dive layer when implementation detail is needed

​

Not Sure Where to Start?