How one can Construct a Value-Conscious LLM Routing System with NadirClaw Utilizing Native Immediate Classification and Gemini Mannequin Switching

if proxy_alive():
   print("n[10] Blended 10-prompt workload…")
   workload = [
       "Capital of France?",
       "Read foo.py",
       "Type hint for a list of dicts",
       "Lowercase: HELLO",
       "One-sentence summary of REST",
       "Refactor a callback chain into async/await with proper error handling",
       "Design a sharded multi-region key-value store with linearizable reads",
       "Analyze the asymptotic complexity of this code and prove the bound rigorously",
       "Debug why our gRPC stream stalls when the client TCP window saturates",
       "Compare and contrast B-trees and LSM-trees for write-heavy workloads",
   ]
   runs = []
   consumer = OpenAI(base_url=f"http://localhost:{PORT}/v1", api_key="native")
   for p in workload:
       t0 = time.time()
       strive:
           r = consumer.chat.completions.create(
               mannequin="auto",
               messages=[{"role": "user", "content": p}],
               max_tokens=140,
           )
           utilization = getattr(r, "utilization", None)
           runs.append({
               "immediate": p[:55],
               "mannequin": r.mannequin,
               "latency_s": spherical(time.time() - t0, 2),
               "in_tok": getattr(utilization, "prompt_tokens", 0) if utilization else 0,
               "out_tok": getattr(utilization, "completion_tokens", 0) if utilization else 0,
           })
       besides Exception as e:
           runs.append({"immediate": p[:55], "mannequin": "ERROR",
                        "latency_s": None, "in_tok": 0, "out_tok": 0,
                        "error": str(e)[:80]})
   rdf = pd.DataFrame(runs)
   print(rdf.to_string(index=False))
   PRICE = {
       "flash": {"in": 0.30 / 1e6, "out": 2.50 / 1e6},
       "professional":   {"in": 1.25 / 1e6, "out": 10.0 / 1e6},
   }
   def price_for(model_str, in_t, out_t):
       m = (model_str or "").decrease()
       tier = "flash" if "flash" in m else "professional"
       return in_t * PRICE[tier]["in"] + out_t * PRICE[tier]["out"]
   cost_routed = sum(price_for(r["model"], r["in_tok"], r["out_tok"]) for r in runs)
   cost_no_route = sum(price_for("gemini-2.5-pro", r["in_tok"], r["out_tok"]) for r in runs)
   print(f"n[10] Value (NadirClaw routed)        : ${cost_routed:.6f}")
   print(f"     Value (always-Professional baseline)     : ${cost_no_route:.6f}")
   if cost_no_route > 0:
       print(f"     Estimated financial savings on this run  : "
             f"{(1 - cost_routed/cost_no_route) * 100:.1f}%")
print("n[11] `nadirclaw report` (parses the JSONL request log):")
rep = subprocess.run(["nadirclaw", "report"], capture_output=True, textual content=True, timeout=60)
print(rep.stdout or rep.stderr)
if proxy_alive():
   print("n[12] Stopping the proxy…")
   strive:
       if hasattr(os, "killpg"):
           os.killpg(os.getpgid(server_proc.pid), sign.SIGTERM)
       else:
           server_proc.terminate()
       server_proc.wait(timeout=10)
   besides Exception:
       strive:
           server_proc.kill()
       besides Exception:
           go
   print("    ✓ proxy stopped.")
print("nDone. 🎉")