nlproxy 1.2.1

High-performance prompt compression and LLM proxy SDK by IntelliDeep

nlproxy-python: High-Performance Python SDK (v1.0.0)

This branch contains the high-performance Python SDK bindings for nlproxy, compiled from our Rust core engine using Maturin and PyO3.

It provides the speed of native Rust execution (sub-3ms latency, 5000+ rps) wrapped in an easy-to-use Python interface.

---

📦 Installation

To install this high-performance version of the SDK directly from GitHub:

# Install directly from the sdk-python branch
pip install git+https://github.com/intellideep/nlproxy.git@sdk-python

Note: This will automatically build and compile the Rust core on your local system. Ensure you have the Rust toolchain (rustc and cargo) installed.

---

💻 Code Example

from nlproxy import init_engine, ensure_models_ready, compress_prompt, run_unified_pipeline, CompressRequest, CompressUnifiedRequest

# 1. Download models automatically if needed
ensure_models_ready("models")

# 2. Initialize the Candle embedding engine with local models
init_engine(
    "models/all-MiniLM-L6-v2/model.safetensors",
    "models/all-MiniLM-L6-v2/config.json",
    "models/all-MiniLM-L6-v2/tokenizer.json"
)

# 3. Compress prompt and redact PII
req = CompressRequest(
    text="The database server is located at 192.168.1.150.",
    mode="general",
    aggressiveness=0.5
)
res = compress_prompt(req)

print("Shielded Prompt:", res.processed_text)
# "The database server is located at __PROT_82736284__."

print("PII Placeholders:", res.placeholders)
# { "__PROT_82736284__": "192.168.1.150" }

Run Unified Orchestrated Pipeline

Execute a complete secure LLM pipeline including firewall checks, Redis semantic cache lookup, prompt compression, upstream LLM execution, and post-LLM verification:

# Create pipeline request
pipeline_req = CompressUnifiedRequest(
    prompt="Connect to database and extract tables",
    domain="general",
    aggressiveness=0.0,
    provider="gemini",
    model="gemini-1.5-pro",
    bypass_cache=False,
    check_firewall=True
)

# Run pipeline
pipeline_res = run_unified_pipeline(pipeline_req)

print("Allowed:", pipeline_res.allowed)
print("Response:", pipeline_res.final_response)
print("Latency:", pipeline_res.latency_ms, "ms")

---

🏢 Authors & Cofounders

This SDK is developed and maintained exclusively by IntelliDeep Labs.

• B-GUST (Co-founder / Lead Developer): github.com/B-GUST
• luiserb (Co-founder / Architect): github.com/luiserb

For attributions and licenses of integrated open-source components, see credits.md.

---

© 2026 IntelliDeep Labs. All rights reserved.