Prompt routing,
crystallized.

# Launch the daemon, then route over Base64 stdin/stdout IPC
# (Windows/MSVC path shown — Make/Ninja builds: aenea-pinta-engine/build/cittern_daemon)
import base64, json, subprocess

daemon = subprocess.Popen(
    [r"aenea-pinta-engine\build\Release\cittern_daemon.exe"],
    stdin=subprocess.PIPE, stdout=subprocess.PIPE,
    text=True, encoding="utf-8")

while "READY" not in daemon.stdout.readline():
    pass                                  # engine warm-up signal

prompt = "Write a Python script for validating user schema in FastAPI."
daemon.stdin.write(base64.b64encode(prompt.encode()).decode() + "\n")
daemon.stdin.flush()

r = json.loads(daemon.stdout.readline())
print(f"Tier: {r['tier']} | Latency: {r['engine_latency_ms']:.2f}ms")
# → Tier: C | Latency: 54.48ms

# Direct ONNX graph inference in Python
import numpy as np, onnxruntime as ort
from transformers import AutoTokenizer

session = ort.InferenceSession("cittern_1_50m.onnx",
                               providers=["CPUExecutionProvider"])
tok = AutoTokenizer.from_pretrained("JamesQuartz/aenea-pinta-1.0")

inputs = tok("Derive the thermodynamic efficiency of an ideal Diesel cycle.",
             return_tensors="np", padding="max_length",
             max_length=512, truncation=True)

logits = session.run(None, {"input_ids": inputs["input_ids"].astype(np.int64)})[0]
print(f"Logits shape: {logits.shape}")

# Inspect parameters or load state dicts into PyTorch
from safetensors.torch import load_file

state_dict = load_file("model.safetensors")
print(f"Loaded {len(state_dict)} tensors.")

total = sum(p.numel() for p in state_dict.values())
print(f"Total parameters: {total / 1e6:.2f}M")
# → Total parameters: 52.41M