sovereign-sieve 1.1.0

Zero-dependency standalone Prose Tax token optimization utility for the Sovereign Systems SDK

sovereign-sieve

Zero-dependency standalone Prose Tax token optimization utility.

sovereign-sieve extracts the payload-cleansing mechanics from the Sovereign Systems SDK into a pure, synchronous micro-library with no web framework or ML library dependencies. Use it in scripts, batch pipelines, and offline data-processing workers to strip conversational boilerplate and measure immediate token savings — all on local silicon.

---

Installation

pip install sovereign-sieve

Or, within a uv workspace:

uv add sovereign-sieve

---

Quick Start

pure_sieve — drop-in string cleaner

from sovereign_sieve import pure_sieve

raw = "Hello! Please just help me analyze this dataset."
clean = pure_sieve(raw)
print(clean)
# → "! help me analyze this dataset."

sieve_with_metrics — cleaner + FinOps telemetry

from sovereign_sieve import sieve_with_metrics

result = sieve_with_metrics(
    "Hi! I hope this helps. Please just run the pipeline."
)
print(result.text)
# → "! helps. run the pipeline."

print(result.raw_token_count)        # estimated tokens before sieve
print(result.optimized_token_count)  # estimated tokens after sieve
print(result.tax_savings_percentage) # e.g. 66.6667 (%)

Offline batch pipeline

from sovereign_sieve import pure_sieve

records = load_jsonl("prompts.jsonl")
cleaned = [pure_sieve(r["text"]) for r in records]
save_jsonl("prompts_clean.jsonl", cleaned)

---

What gets stripped

Category	Examples stripped	Examples preserved
Greeting tokens	hi, hello, hey, greetings	hi-fi, hello-world
Hedging adverbs	please, just, simply, actually, basically, probably	just-in-time, simply-typed
Affirmation filler	of course, certainly, absolutely, sure (standalone)	make sure to, be sure to, certainly-not
Preamble phrases	I hope this, I hope that, I hope you	trailing sentence content preserved
Degenerate markdown	****, _____ (4+ bare markers)	**bold**, *italic*, ***bold-italic***
Duplicate headings	consecutive identical ## Heading\n## Heading	headings differing in case or level

---

SieveOutput fields

from sovereign_sieve import SieveOutput

result: SieveOutput = sieve_with_metrics("Hello! Please help me now.")
result.text                  # str  — minimized payload
result.raw_token_count       # int  — byte-density token estimate before sieve
result.optimized_token_count # int  — byte-density token estimate after sieve
result.tax_savings_percentage # float — reduction percentage, 4 decimal places

Token counts use the UTF-8 byte-density heuristic (÷ 4), matching the average density of the cl100k_base tokenizer across mixed English/code content. No external tokenizer is invoked.

---

Key manager usage (sovereign-core)

For cryptographic signing of sieved payloads, use SovereignGateway from sovereign-core:

import asyncio
from sovereign_core.gateway import SovereignGateway

async def main():
    gateway = SovereignGateway(signing_key=".keys/sovereign_identity.pem")
    result = await gateway.sieve_and_sign("Hello! Please just help me.")
    print(result.content)          # minimized text
    print(result.receipt["signature"])  # Ed25519 signature

asyncio.run(main())

---

License

MIT