On this page
01
Point ChatOpenAI at Cloptima
LangChain's ChatOpenAI accepts a base URL and default headers, so chains and agents route through Cloptima with no structural change.
python
from langchain_openai import ChatOpenAI
llm = ChatOpenAI(
model="gpt-5.4-mini",
base_url="https://api.cloptima.ai/v1/ai",
api_key="clop_vk_dPXO67p…",
default_headers={"x-cloptima-team": "research", "x-cloptima-app": "agent"},
)02
Track runs, not just requests
A single user action can fan out into many model calls. Pass x-cloptima-session-id and x-cloptima-run-id (and agent-session/agent-run ids) per invocation so cost rolls up to the chain, tool, and agent run — not a flat request count.
03
Control expensive workflows
Apply stricter budgets to experimental agents, long-running tools, high-context prompts, and retry-heavy paths.
04
Analyze cost drivers
Break spend down by model, chain, tool, team, app, and environment to find the workflows that need prompt, routing, or architecture changes.