Managing LangChain Agents with a Control Plane

LangChain + Control Plane

LangChain excels at building agent chains — sequences of LLM calls, tool uses, and retrieval steps. But as chains grow in complexity, so do costs, debugging difficulty, and compliance requirements. A control plane adds the operational layer LangChain does not provide.

Integration Code

from langchain_openai import ChatOpenAI
from langchain.agents import AgentExecutor, create_openai_tools_agent
from langchain_core.prompts import ChatPromptTemplate

# Point LangChain to the Dobby Gateway
llm = ChatOpenAI(
    model="gpt-4o",
    openai_api_base="https://dobby-ai.com/api/v1/gateway",
    openai_api_key="gk_svc_your_service_key"
)

# Use LangChain as usual — the Gateway handles the rest
prompt = ChatPromptTemplate.from_messages([
    ("system", "You are a helpful research assistant."),
    ("human", "{input}"),
])

agent = create_openai_tools_agent(llm, tools=[], prompt=prompt)
executor = AgentExecutor(agent=agent, tools=[], verbose=True)

result = executor.invoke({"input": "Analyze Q1 sales data"})

Multi-Provider Routing

A major advantage of routing through a gateway is provider flexibility. Your LangChain chain can use GPT-4o for complex reasoning, Claude for long documents, and Gemini for simple classification — all through the same endpoint. Switch models by changing a parameter, not your code.

# Same gateway, different models
llm_reasoning = ChatOpenAI(model="gpt-4o", openai_api_base=GATEWAY_URL, openai_api_key=KEY)
llm_documents = ChatOpenAI(model="claude-sonnet-4-20250514", openai_api_base=GATEWAY_URL, openai_api_key=KEY)
llm_classify  = ChatOpenAI(model="gemini-2.5-flash", openai_api_base=GATEWAY_URL, openai_api_key=KEY)

# Each call is tracked separately with provider-specific cost

What You Get

• Per-chain cost breakdown — see how much each step in your chain costs
• Provider comparison — which model delivers the best cost-to-quality ratio
• Real-time monitoring — watch chain execution live as requests flow through
• Budget enforcement — chains stop when they hit cost limits, not after
• Audit trail — every LLM call in every chain, queryable and exportable

LangSmith + Dobby

LangSmith provides detailed chain tracing. Dobby provides cost governance and multi-framework management. They are complementary — use LangSmith for debugging chain logic, use Dobby for cost control, approvals, and fleet management across all your agents (not just LangChain).