more

cmd/needle-sidecar/server.py

@@ -8,6 +8,8 @@ import os
 import sys
 import json
 import time
+import re
+import hashlib
 from http.server import ThreadingHTTPServer, BaseHTTPRequestHandler
 import needle
 
@@ -50,6 +52,31 @@ def escalate_backend_outage(service: str, error_message: str):
     """Escalate severe backend 5xx crash or unhandled panic to on-call."""
     return {"action": "escalate_outage", "service": service, "error": error_message}
 
+@needle.tool
+def report_unclassified(service: str, summary: str):
+    """Fallback when the event matches none of the other categories: retries, cancellations, restarts, OOM, health checks, or anything unclear."""
+    return {"action": "unclassified", "service": service, "summary": summary}
+
+# --- Redaction: nothing PII-bearing should reach the model or the cache ---
+_REDACTIONS = [
+    (re.compile(r"[\w.+-]+@[\w-]+\.[\w.-]+"), "<email>"),
+    (re.compile(r"\b(?:\d{1,3}\.){3}\d{1,3}\b"), "<ip>"),
+    (re.compile(r"\b[0-9a-fA-F]{8}-(?:[0-9a-fA-F]{4}-){3}[0-9a-fA-F]{12}\b"), "<uuid>"),
+    (re.compile(r"\beyJ[\w-]+\.[\w-]+\.[\w-]+\b"), "<jwt>"),
+    (re.compile(r"\b[0-9a-fA-F]{16,}\b"), "<hex>"),
+    (re.compile(r"([?&][\w.-]+=)[^&\s|]+"), r"\1<v>"),
+]
+
+def redact(text: str) -> str:
+    for rx, sub in _REDACTIONS:
+        text = rx.sub(sub, text)
+    return text[:1000]
+
+# Cache keyed on the redacted text: repeated signatures cost nothing.
+CACHE_MAX = int(os.environ.get("NEEDLE_CACHE_MAX", 2000))
+_cache = {}
+_stats = {"hits": 0, "misses": 0}
+
 # Initialize Needle Agent once
 print(f"Loading Needle 3 model (Cactus-Compute/needle3)...")
 t0 = time.perf_counter()
@@ -60,7 +87,8 @@ agent = needle.Needle(tools=[
     log_missing_resource_404,
     report_database_error,
     report_auth_failure,
-    escalate_backend_outage
+    escalate_backend_outage,
+    report_unclassified
 ])
 print(f"Needle 3 loaded in {(time.perf_counter() - t0):.2f}s")
 
@@ -86,6 +114,7 @@ class NeedleHandler(BaseHTTPRequestHandler):
                 "status": "ok",
                 "available": True,
                 "model": "Cactus-Compute/needle3",
+                "cache": {**_stats, "size": len(_cache)},
                 "tools": [
                     "client_js_crash",
                     "security_alert",
@@ -93,7 +122,8 @@ class NeedleHandler(BaseHTTPRequestHandler):
                     "log_404",
                     "database_error",
                     "auth_failure",
-                    "escalate_outage"
+                    "escalate_outage",
+                    "unclassified"
                 ]
             }
             self.wfile.write(json.dumps(resp).encode("utf-8"))
@@ -142,6 +172,20 @@ class NeedleHandler(BaseHTTPRequestHandler):
                 self.wfile.write(json.dumps({"error": "Missing 'text' or telemetry fields"}).encode("utf-8"))
                 return
 
+            text = redact(text)
+            key = hashlib.sha256(text.encode()).hexdigest()
+            cached = _cache.get(key)
+            if cached is not None:
+                _stats["hits"] += 1
+                out = {**cached, "cached": True}
+                self.send_response(200)
+                self._send_cors()
+                self.send_header("Content-Type", "application/json")
+                self.end_headers()
+                self.wfile.write(json.dumps(out).encode("utf-8"))
+                return
+            _stats["misses"] += 1
+
             t_start = time.perf_counter()
             agent.reset()
             res = agent.run(text, strict=False)
@@ -166,6 +210,10 @@ class NeedleHandler(BaseHTTPRequestHandler):
                 "model": "Cactus-Compute/needle3"
             }
 
+            if len(_cache) >= CACHE_MAX:
+                _cache.pop(next(iter(_cache)))
+            _cache[key] = out
+
             self.send_response(200)
             self._send_cors()
             self.send_header("Content-Type", "application/json")