Skip to main content
The Agent class supports context compression to handle large amounts of information efficiently. This is particularly useful when dealing with long conversation histories or large documents that might exceed the model’s context window.

​
Compression Strategies

Upsonic offers different compression strategies configured via compression_strategy:
  1. none (Default): No compression.
  2. simple: Basic whitespace removal and middle truncation if length exceeds limit.
  3. llmlingua: Advanced compression using the LLMLingua library (requires pip install llmlingua).

​
Simple Compression

Use simple compression for basic optimization without additional dependencies. 
from upsonic import Agent, Task

# Configure simple compression
agent = Agent(
    model="openai/gpt-4o",
    compression_strategy="simple",
    compression_settings={
        "max_length": 1000  # distinct characters
    }
)

# Task with potentially long context
long_text = "..." * 10000
task = Task(f"Summarize this text: {long_text}")

result = agent.do(task)
print(result)

​
LLMLingua Compression

Use LLMLingua for smart, semantic compression that preserves key information. 
from upsonic import Agent, Task

# Configure smart compression
agent = Agent(
    model="openai/gpt-4o",
    compression_strategy="llmlingua",
    compression_settings={
        "ratio": 0.5,           # Keep 50% of the tokens
        "instruction": "Keep the financial details",
        "model_name": "microsoft/llmlingua-2-xlm-roberta-large-meetingbank"
    }
)

task = Task("Analyze the provided meeting transcript.")
result = agent.do(task)
print(result)
Using llmlingua requires installing the package: pip install llmlingua. The agent will perform the compression locally before sending the prompt to the LLM.