Recommendations guide
Token compression for AI recommendations
LLM-based recommendation engines send user profiles, product catalogs, and browsing history with every request. Compression removes irrelevant products and attributes from the context.
How recommendations use LLMs
Modern recommendation engines use LLMs to generate personalized product suggestions. A typical request includes: the user's previous purchases (3-10 items, 500-2000 tokens), browsing history (5-20 pages, 300-1500 tokens), demographic data (50-200 tokens), and the product catalog subset (500-3000 tokens). Total: 1,350-6,700 tokens per request.
Compressing recommendation context
from supercompress import Compressor
comp = Compressor()
def recommend(user_profile, product_catalog, current_query):
context = f"User: {user_profile}\nCatalog: {product_catalog}"
result = comp.compress(context, current_query)
# Only user attributes and products relevant to the query remain
return llm.generate(
f"Recommend products based on: {result.compressed_text}"
)Frequently asked questions
Does compression improve recommendation quality?
Often yes. Removing irrelevant products helps the LLM focus on the right candidates.
Can I use this even without an LLM for recommendations?
Yes. SuperCompress works as a pre-processing step regardless of the recommendation algorithm.
Build with less context
Put compression in front of your next LLM call.
Use the hosted API or run SuperCompress locally. Keep the evidence, drop the token waste, and measure the savings before it reaches your model.