preduce 0.3.0

Compress LLM prompts 50%+ while preserving meaning. Lightweight API client.

preduce

Compress LLM prompts 50%+ while preserving meaning. Cut your API costs in half with one line of code.

Install

pip install preduce

Quick Start

from preduce import compress

result = compress("Your verbose text here...", api_key="your-key")

print(result.compressed_text)      # compressed output
print(result.category)             # GENERAL, FINANCIAL, MEDICAL, CODE, ACADEMIC, SUPPORT, or MARKETING
print(result.token_reduction_pct)  # e.g. 52.3
print(result.stats)                # full reduction metrics

Category Override

result = compress(
    "Total revenue for fiscal year 2025 was $847,300,000, "
    "representing an increase of 22.4% year-over-year.",
    api_key="your-key",
    category="FINANCIAL",
)
# → "Total rev FY2025 $847.3M, +22.4% YoY."

Categories

Category	What it compresses	Example
GENERAL	Business docs, emails, reports	"we are pleased to announce" → removed
FINANCIAL	Earnings, metrics, filings	"$847,300,000" → "$847.3M"
MEDICAL	Clinical notes, discharge summaries	"type 2 diabetes mellitus" → "T2DM"
CODE	Preserves code, compresses prose	Code untouched, comments compressed
ACADEMIC	Research papers, dissertations	"statistically significant" → "sig."
SUPPORT	Support tickets, help desk	"thank you for contacting" → "thanks for contacting"
MARKETING	Marketing copy, promotions	"limited time offer" → "limited time"

Response

result.compressed_text   # str — the compressed text
result.category          # str — detected or overridden category
result.original_tokens   # int — estimated original token count
result.compressed_tokens # int — estimated compressed token count
result.confidence_pct    # dict — confidence % per category (e.g. {"FINANCIAL": 85.2, "GENERAL": 14.8, ...})
result.token_reduction_pct  # float — percentage reduction
result.char_reduction_pct   # float — percentage reduction
result.stats             # dict — full metrics
result.to_dict()         # dict — serialize everything

Confidence Scores

The API returns confidence percentages for each category, useful for debugging classification:

result = compress("Revenue increased 15% this quarter.", api_key="your-key")
print(result.confidence_pct)
# {"FINANCIAL": 78.5, "GENERAL": 12.3, "MARKETING": 9.2, ...}
print(result.category)  # "FINANCIAL"

Error Handling

from preduce import compress

try:
    result = compress("text", api_key="invalid-key")
except PermissionError:
    print("Bad API key or quota exceeded")
except ValueError:
    print("Invalid input (empty text or bad category)")
except RuntimeError:
    print("API error")

Get an API Key

Visit preduce.dev to get your API key.

License

MIT