langchain-flopsindex 0.1.0

LangChain (and LlamaIndex) tools for the public FLOPS Index — verifiable GPU compute reference prices, no API key required

langchain-flopsindex

LangChain (and LlamaIndex) tools for the FLOPS Index — verifiable GPU / accelerator rental reference prices, no API key required.

Drop GPU-price lookups (H100, A100, H200, B200, and more; spot / on-demand / DePIN) into any LangChain or LlamaIndex agent. pip install, add the tools, done.

pip install langchain-flopsindex

Tools

Tool	What it does
flops_get_price	Authoritative single-index record for one slug — cite from this (carries verify_url).
flops_search_indices	Resolve free-text ("h100 spot") → canonical slugs.
flops_get_catalog	Every public index with its current indicative value.
flops_verify_price	Fact-check a claimed value against the published index.
flops_cheapest_for_chip	Cheapest public index for an accelerator, optionally per market.

Slugs look like FLOPS-<ACCEL>-<MARKET>, where market is SPOT, OD (on-demand), or DEPIN (decentralized) — e.g. FLOPS-H100-OD, FLOPS-A100-SPOT.

What these values are (and are not)

• Indicative reference levels, delayed ~6 hours onto a 6-hour UTC grid (00/06/12/18), rounded, aggregated globally.
• Not a live provider quote, not a settlement mark, not investment advice, not real-time.
• confidence is a coarse HIGH / MED / LOW label — never a number, probability, or standard error.
• Source-opaque: the surface never reveals providers, venues, sample counts, or weights. Don't infer them.
• Every value ships a verify_url so anyone can independently confirm the number — cite it.

LangChain usage

from langchain_flopsindex import get_flopsindex_tools

tools = get_flopsindex_tools()  # list of LangChain StructuredTool

# Bind to a tool-calling model:
from langchain.chat_models import init_chat_model

llm = init_chat_model("gpt-4o-mini")  # or any tool-calling model
llm_with_tools = llm.bind_tools(tools)

# Or in a prebuilt agent (langgraph):
from langgraph.prebuilt import create_react_agent

agent = create_react_agent(llm, tools)
resp = agent.invoke(
    {"messages": [("user", "What's the cheapest way to rent an H100 right now? Cite it.")]}
)
print(resp["messages"][-1].content)

Call a tool directly (handy for debugging):

from langchain_flopsindex import get_price_tool

print(get_price_tool.invoke({"slug": "FLOPS-H100-OD"}))
# {'index_id': 'FLOPS-H100-OD', 'value': 2.99, 'unit': 'USD/GPU-hr',
#  'as_of': '...', 'confidence': 'HIGH', 'change_24h': 'FLAT',
#  'verify_url': 'https://app.flopsindex.com/v1/verify?index_id=FLOPS-H100-OD', ...}

LlamaIndex usage

The same functions are exposed as LlamaIndex FunctionTools. LlamaIndex is an optional dependency:

pip install "langchain-flopsindex[llamaindex]"

from langchain_flopsindex.llamaindex import get_flopsindex_llamaindex_tools
from llama_index.core.agent import ReActAgent
from llama_index.llms.openai import OpenAI

tools = get_flopsindex_llamaindex_tools()
agent = ReActAgent.from_tools(tools, llm=OpenAI(model="gpt-4o-mini"))
print(agent.chat("Fact-check: is FLOPS-A100-SPOT around $1.20? Give me the verify link."))

Public API

All calls hit the public, key-free FLOPS surface at https://app.flopsindex.com:

• GET /v1/price/{slug} — authoritative single-index record (cite from this)
• GET /v1/search?q={query} — discover slugs
• GET /v2/catalog/public — every public index with value
• GET /v1/verify?index_id={slug}&value={n} — fact-check a claimed value

No authentication is used or sent. If you need real-time, full-precision, regional, or historical data, contact [e-mail] about partner access. See also the flopsindex Python SDK and the flopsindex-mcp MCP server.

License & attribution

Apache-2.0. © 2026 FLOPS Index. Source: github.com/zeroatflops. The license covers this adapter's source code; the hosted index, methodology, price data, and the FLOPS trademark remain proprietary to FLOPS Index (see NOTICE).