Studio: add option to disable the in-memory API monitor

studio/backend/core/inference/api_monitor.py

@@ -5,6 +5,7 @@
 
 from __future__ import annotations
 
+import os
 import threading
 import time
 import uuid
@@ -18,6 +19,17 @@
 _MAX_REPLY_CHARS = 12000
 _PREVIEW_CHARS = 360
 
+# Opt-in kill switch for the in-memory API monitor. Users who run Studio purely
+# as an inference API and handle logging/telemetry elsewhere can set this to turn
+# the monitor into a no-op (nothing is recorded and the Monitor view stays empty).
+# Off by default, so existing behaviour is unchanged.
+_DISABLE_ENV = "UNSLOTH_STUDIO_DISABLE_API_MONITOR"
+_TRUE_VALUES = frozenset({"1", "true", "yes", "on"})
+
+
+def _api_monitor_disabled() -> bool:
+    return os.environ.get(_DISABLE_ENV, "").strip().lower() in _TRUE_VALUES
+
 
 def _trim(text: Optional[str], limit: int) -> str:
     if not text:
@@ -93,10 +105,16 @@ def snapshot(self, *, include_details: bool = True) -> dict[str, Any]:
 
 
 class ApiMonitor:
-    def __init__(self, max_entries: int = _MAX_ENTRIES):
+    def __init__(
+        self,
+        max_entries: int = _MAX_ENTRIES,
+        *,
+        enabled: bool = True,
+    ):
         self._entries: deque[ApiMonitorEntry] = deque()
         self._max_entries = max(0, max_entries)
         self._lock = threading.Lock()
+        self._enabled = enabled
 
     def start(
         self,
@@ -108,6 +126,11 @@ def start(
         context_length: Optional[int] = None,
         subject: Optional[str] = None,
     ) -> str:
+        # Disabled monitor is a no-op: return a falsy id so every downstream
+        # mutator (append_reply/set_reply/set_usage/finish/fail) short-circuits
+        # on its `if not entry_id` guard and no entry is ever recorded.
+        if not self._enabled:
+            return ""
         now = time.time()
         entry = ApiMonitorEntry(
             id = f"apireq_{uuid.uuid4().hex[:12]}",
@@ -292,4 +315,4 @@ def _trim_terminal_locked(self) -> None:
         self._entries = kept
 
 
-api_monitor = ApiMonitor()
+api_monitor = ApiMonitor(enabled = not _api_monitor_disabled())

studio/backend/tests/test_api_monitor.py

@@ -258,3 +258,61 @@ def test_api_monitor_append_reply_exact_cap_then_more_marks_truncated():
     monitor.append_reply(entry_id, "y")
     reply = monitor.snapshot()[0]["reply"]
     assert len(reply) == m._MAX_REPLY_CHARS and reply.endswith("...")
+
+
+def test_api_monitor_enabled_by_default_records():
+    # Guard against accidentally flipping the default to disabled.
+    monitor = ApiMonitor(max_entries = 3)
+    entry_id = monitor.start(
+        endpoint = "/v1/chat/completions",
+        method = "POST",
+        model = "local-model",
+        prompt = "user: hello",
+    )
+    assert entry_id
+    assert len(monitor.snapshot()) == 1
+
+
+def test_api_monitor_disabled_is_noop():
+    monitor = ApiMonitor(max_entries = 3, enabled = False)
+
+    entry_id = monitor.start(
+        endpoint = "/v1/chat/completions",
+        method = "POST",
+        model = "local-model",
+        prompt = "user: hello",
+        context_length = 100,
+    )
+    # A disabled monitor returns a falsy id and records nothing.
+    assert entry_id == ""
+
+    # Every mutator must be a safe no-op on the falsy id.
+    monitor.append_reply(entry_id, "hi")
+    monitor.set_reply(entry_id, "hi")
+    monitor.set_usage(entry_id, prompt_tokens = 4, completion_tokens = 6)
+    monitor.finish(entry_id)
+    monitor.fail(entry_id, "boom")
+
+    assert monitor.snapshot() == []
+    assert monitor.active_count() == 0
+    assert monitor.get(entry_id) is None
+
+
+def test_api_monitor_disable_env_var_truthy(monkeypatch):
+    import core.inference.api_monitor as m
+    for value in ("1", "true", "yes", "on", "TRUE", "On", " yes "):
+        monkeypatch.setenv(m._DISABLE_ENV, value)
+        assert m._api_monitor_disabled() is True, value
+
+
+def test_api_monitor_disable_env_var_falsy(monkeypatch):
+    import core.inference.api_monitor as m
+    for value in ("", "0", "false", "no", "off", "disabled"):
+        monkeypatch.setenv(m._DISABLE_ENV, value)
+        assert m._api_monitor_disabled() is False, value
+
+
+def test_api_monitor_disable_env_var_unset(monkeypatch):
+    import core.inference.api_monitor as m
+    monkeypatch.delenv(m._DISABLE_ENV, raising = False)
+    assert m._api_monitor_disabled() is False