CAELUM

CAELUM Cloud

A shared cache for all your servers, cost per customer, and budget alerts.

CAELUM Cloud is optional. Without it, CAELUM works fully on your machine. With a CAELUM key, the same SDK also:

  • shares its cache between all your servers, including short-lived serverless instances (Vercel, AWS Lambda, Cloud Run) where a local cache file disappears after each request;
  • shows requests, cache hits, money saved and money spent in a dashboard, per day, per model and per end customer;
  • sends budget alerts before a surprise provider bill.
from caelum import OpenAI

client = OpenAI(cache=True, caelum_key="sk-caelum-...")  # or set CAELUM_API_KEY

Your AI calls never go through CAELUM Cloud: the provider is always called directly, with your own API key, which never leaves your application.

Cost per customer

Attribute requests to one of your own customers to see what each of them costs you:

import caelum

with caelum.customer("customer-42"):
    client.chat.completions.create(model="gpt-4o-mini", messages=[...])

It works with threads and asyncio (each task keeps its own value). Use an internal id, not personal data such as an email or a name: it is sent to CAELUM Cloud as a label.

What is sent

Sent to CAELUM Cloud Never sent
Model, token counts, estimated cost and savings, cache hit or miss, latency, customer label Your provider API keys
With cache=True: cached answers, in your own isolated account Prompts in the metrics

Each account's cache is isolated: another CAELUM customer can never read your cached answers. Only a hash of your CAELUM key is stored on our side.

CAELUM Cloud never breaks your app

  • A cache lookup waits at most 0.5 s for the Cloud; metrics are sent in the background.
  • If the Cloud is slow or unreachable, it is switched off for a minute and the app goes on with its local cache.
  • An invalid or revoked key logs one warning and disables the Cloud for that process.
  • Over the monthly token limit of your plan, the shared cache pauses; metrics are still recorded and your app keeps working.

Plans

Plan Price Tokens per month
Free $0 1 million
Starter $49 (29,500 FCFA) per 30 days, prepaid (card or Mobile Money, no automatic charge) 10 million

Only tokens actually billed by your provider count (cache hits are free). See Pricing.

← ReliabilityAPI reference →