1. Install (30 seconds)
pip install "caelum-ai[openai]" # OpenAI
pip install "caelum-ai[anthropic]" # Anthropic
pip install "caelum-ai[openai,anthropic,rich]" # both, with pretty metrics
Python 3.10 or later. No server, no database, no account.
2. Change one line
from caelum import OpenAI # was: from openai import OpenAI
client = OpenAI(cache=True) # reads OPENAI_API_KEY, like the SDK
response = client.chat.completions.create(
model="gpt-4o-mini",
messages=[{"role": "user", "content": "How do I return an item?"}],
)
print(response.choices[0].message.content)
Run the same request twice: the second answer comes from the local cache, instantly and for free.
3. See what you saved
client.metrics.print_summary()
CAELUM Local Metrics
Requests: 2
Cache hits: 1 (50%)
Cost saved: $ 0.000007
Cost spent: $ 0.000007
Avg latency: 310ms
p95 latency: 600ms
4. Go further
import caelum
from caelum import OpenAI
client = OpenAI(
cache=True,
semantic_cache=True, # paraphrases reuse answers (sends the question to the embeddings API)
retry=True,
)
response = client.chat.completions.create(
model="gpt-4o-mini",
messages=[{"role": "user", "content": "How can I send back a product?"}],
)
caelum.cache_type(response) # "exact", "semantic" or None
- Caching — exact and semantic cache.
- Routing — send requests to cheaper models with simple rules.
- Reliability — retries and multi-provider fallback.