Track what your LlamaIndex pipelines cost

pip install focxle llama-index-llms-openai

focxle has no dependencies, so nothing enters your environment and nothing can conflict with the versions you already pin.

focxle.init() goes as early as you can put it, before the clients are built. Naming agents is what makes the per agent number exist at all.

import focxle
focxle.init()

from llama_index.llms.openai import OpenAI

llm = OpenAI(model="gpt-5.5")

with focxle.agent("research"):
    llm.complete("what changed in the filing")

Printed when the process exits, or on demand with focxle.report():

  focxle spend                                    $16.17
  --------------------------------------------------------
  research                            $16.17  (203 calls)
      openai                                       $14.02
      api.firecrawl.dev                             $2.15

Model calls and HTTP APIs in the same number, per agent. That second part is what no LLM-only tool can show you: the search, scraping and embedding calls your agent makes are real money and they appear on a different bill.

Embedding calls are counted too, since they are ordinary HTTP requests to a priced host. Retrieval-heavy pipelines usually spend more on embeddings than anyone expects.

Query engines, chat engines and agents all route through the same client, so none of them need separate handling.

It does not block anything. Out of the box it only measures, and it shows what a spending cap would have stopped so you can decide whether you want one before paying for one.

It is not a proxy. Your traffic never routes through us, so there is no new dependency on your live path and we never hold your prompts. It makes no network calls at all on the free path, which the test suite proves by deleting socket.socket and running anyway.

It fails open. If anything inside it breaks, your call still goes through and the failure is counted and printed at the end.