diff --git a/studio/backend/core/inference/api_monitor.py b/studio/backend/core/inference/api_monitor.py
index e4aad4ade2..8b3cd02d85 100644
--- a/studio/backend/core/inference/api_monitor.py
+++ b/studio/backend/core/inference/api_monitor.py
@@ -41,6 +41,10 @@ class ApiMonitorEntry:
     started_at: float
     updated_at: float
     subject: Optional[str] = None
+    # True when the caller used an sk-unsloth key rather than a UI session. The
+    # floating panel only opens itself for these: Studio's own chat goes through
+    # the same endpoints, and popping the monitor open mid-chat is noise.
+    via_api_key: bool = False
     # Monotonic anchors so duration math survives wall-clock steps (NTP).
     started_monotonic: float = 0.0
     finished_monotonic: Optional[float] = None
@@ -70,6 +74,7 @@ class ApiMonitorEntry:
             "endpoint": self.endpoint,
             "method": self.method,
             "model": self.model,
+            "via_api_key": self.via_api_key,
             "prompt_preview": _trim(self.prompt, _PREVIEW_CHARS),
             "reply_preview": _trim(self.reply, _PREVIEW_CHARS),
             "prompt_truncated": len(self.prompt) > _PREVIEW_CHARS,
@@ -107,6 +112,7 @@ class ApiMonitor:
         prompt: str,
         context_length: Optional[int] = None,
         subject: Optional[str] = None,
+        via_api_key: bool = False,
     ) -> str:
         now = time.time()
         entry = ApiMonitorEntry(
@@ -121,6 +127,7 @@ class ApiMonitor:
             started_at = now,
             updated_at = now,
             subject = subject,
+            via_api_key = via_api_key,
             started_monotonic = time.monotonic(),
             context_length = context_length,
         )
diff --git a/studio/backend/routes/inference.py b/studio/backend/routes/inference.py
index d804b3143d..fd1d3cc1cb 100644
--- a/studio/backend/routes/inference.py
+++ b/studio/backend/routes/inference.py
@@ -1754,7 +1754,24 @@ from core.inference.anthropic_compat import (
     AnthropicStreamEmitter,
     AnthropicPassthroughEmitter,
 )
-from auth.authentication import get_current_subject
+from auth.authentication import API_KEY_PREFIX, get_current_subject
+
+
+def _request_used_api_key(request: Any) -> bool:
+    """True when this request authenticated with an sk-unsloth key.
+
+    Studio's own chat hits these same endpoints with a session JWT, so this is
+    what separates "someone is using Unsloth as an API server" from "someone is
+    using Unsloth".
+    """
+    try:
+        header = request.headers.get("authorization") or ""
+    except Exception:
+        return False
+    scheme, _, token = header.partition(" ")
+    return scheme.lower() == "bearer" and token.startswith(API_KEY_PREFIX)
+
+
 from state.tool_approvals import resolve_tool_decision
 
 from core.inference.key_exchange import decrypt_api_key
@@ -7238,6 +7255,7 @@ async def _proxy_to_external_provider(
     if not getattr(request.state, "skip_api_monitor", False):
         monitor_id = api_monitor.start(
             endpoint = request.url.path,
+            via_api_key = _request_used_api_key(request),
             method = request.method,
             model = model,
             prompt = _monitor_prompt_from_messages(payload.messages),
@@ -7798,6 +7816,7 @@ async def openai_chat_completions(
         if not getattr(request.state, "skip_api_monitor", False):
             tts_monitor_id = api_monitor.start(
                 endpoint = request.url.path,
+                via_api_key = _request_used_api_key(request),
                 method = request.method,
                 model = model_label,
                 prompt = _monitor_prompt_from_messages(payload.messages),
@@ -7865,6 +7884,7 @@ async def openai_chat_completions(
         if not getattr(request.state, "skip_api_monitor", False):
             monitor_id = api_monitor.start(
                 endpoint = request.url.path,
+                via_api_key = _request_used_api_key(request),
                 method = request.method,
                 model = model_name,
                 prompt = _monitor_prompt_from_messages(payload.messages),
@@ -8009,6 +8029,7 @@ async def openai_chat_completions(
     if monitor_id is None and not getattr(request.state, "skip_api_monitor", False):
         monitor_id = api_monitor.start(
             endpoint = request.url.path,
+            via_api_key = _request_used_api_key(request),
             method = request.method,
             model = model_name,
             prompt = _monitor_prompt_from_messages(payload.messages),
@@ -10832,6 +10853,7 @@ async def openai_completions(request: Request, current_subject: str = Depends(ge
     prompt_text = _flatten_monitor_prompt(body.get("prompt", ""))
     monitor_id = api_monitor.start(
         endpoint = request.url.path,
+        via_api_key = _request_used_api_key(request),
         method = request.method,
         model = str(body.get("model") or _llama_public_model_id(llama_backend) or "default"),
         prompt = prompt_text,
@@ -11043,6 +11065,7 @@ async def openai_embeddings(request: Request, current_subject: str = Depends(get
     if not getattr(request.state, "skip_api_monitor", False):
         monitor_id = api_monitor.start(
             endpoint = request.url.path,
+            via_api_key = _request_used_api_key(request),
             method = request.method,
             model = str(body.get("model") or _llama_public_model_id(llama_backend) or "default"),
             prompt = prompt_text,
@@ -11637,6 +11660,7 @@ async def _responses_non_streaming(
         monitor_id = api_monitor.start(
             endpoint = getattr(getattr(request, "url", None), "path", "/v1/responses"),
             method = getattr(request, "method", "POST"),
+            via_api_key = _request_used_api_key(request),
             model = payload.model,
             prompt = _monitor_prompt_from_messages(messages),
             context_length = _monitor_context_length(),
@@ -12828,6 +12852,7 @@ async def openai_responses(
         if not getattr(request.state, "skip_api_monitor", False):
             monitor_id = api_monitor.start(
                 endpoint = request.url.path,
+                via_api_key = _request_used_api_key(request),
                 method = request.method,
                 model = payload.model,
                 prompt = _monitor_prompt_from_messages(messages),
@@ -13296,6 +13321,7 @@ async def anthropic_messages(
         monitor_id = api_monitor.start(
             endpoint = getattr(request_url, "path", "/v1/messages"),
             method = getattr(request, "method", "POST"),
+            via_api_key = _request_used_api_key(request),
             model = model_name,
             prompt = _monitor_prompt_from_messages(openai_messages),
             context_length = monitor_context_length,
diff --git a/studio/backend/tests/test_api_monitor.py b/studio/backend/tests/test_api_monitor.py
index a8d0e1e832..c1da17f7bd 100644
--- a/studio/backend/tests/test_api_monitor.py
+++ b/studio/backend/tests/test_api_monitor.py
@@ -289,3 +289,28 @@ def test_api_monitor_clear_is_scoped_to_one_subject():
     # Passing no subject is the explicit "everything" path.
     monitor.clear()
     assert monitor.snapshot(subject = "bob") == []
+
+
+def test_api_monitor_records_whether_the_caller_used_an_api_key():
+    # Studio's own chat hits these endpoints with a session JWT. The floating
+    # panel keys its auto-open off this flag, so mislabelling in-app chat as API
+    # traffic pops the panel over the composer mid-conversation.
+    monitor = ApiMonitor(max_entries = 4)
+    ui = monitor.start(
+        endpoint = "/api/inference/chat",
+        method = "POST",
+        model = "m",
+        prompt = "hi",
+        subject = "u",
+    )
+    api = monitor.start(
+        endpoint = "/v1/chat/completions",
+        method = "POST",
+        model = "m",
+        prompt = "hi",
+        subject = "u",
+        via_api_key = True,
+    )
+    by_id = {entry["id"]: entry for entry in monitor.snapshot(subject = "u")}
+    assert by_id[ui]["via_api_key"] is False
+    assert by_id[api]["via_api_key"] is True
diff --git a/studio/backend/tests/test_openai_auto_switch.py b/studio/backend/tests/test_openai_auto_switch.py
index aa20752aee..b66781375e 100644
--- a/studio/backend/tests/test_openai_auto_switch.py
+++ b/studio/backend/tests/test_openai_auto_switch.py
@@ -4247,3 +4247,20 @@ def test_ambiguous_case_fallback_matches_nothing(monkeypatch):
     settings.set_model_override("/models/foo.gguf", max_seq_length = 1024)
     settings.set_model_override("/models/FOO.gguf", max_seq_length = 8192)
     assert settings.get_model_override("/models/Foo.gguf") == {}
+
+
+def test_request_used_api_key_distinguishes_key_from_session():
+    from auth.authentication import API_KEY_PREFIX
+
+    class _Req:
+        def __init__(self, header):
+            self.headers = {"authorization": header} if header else {}
+
+    assert inference_route._request_used_api_key(_Req(f"Bearer {API_KEY_PREFIX}abc")) is True
+    assert inference_route._request_used_api_key(_Req(f"bearer {API_KEY_PREFIX}abc")) is True
+    assert inference_route._request_used_api_key(_Req("Bearer eyJhbGciOiJIUzI1NiJ9.x")) is False
+    assert inference_route._request_used_api_key(_Req("")) is False
+    assert inference_route._request_used_api_key(_Req(None)) is False
+    # A malformed request object must read as "not an API key", never raise, since
+    # this runs on the hot path of every tracked request.
+    assert inference_route._request_used_api_key(object()) is False
diff --git a/studio/frontend/src/features/api-monitor/api-monitor-overlay.tsx b/studio/frontend/src/features/api-monitor/api-monitor-overlay.tsx
index 88a4915b93..377694d7f5 100644
--- a/studio/frontend/src/features/api-monitor/api-monitor-overlay.tsx
+++ b/studio/frontend/src/features/api-monitor/api-monitor-overlay.tsx
@@ -172,7 +172,12 @@ export function ApiMonitorOverlay(): ReactElement | null {
       return;
     }
     const seen = seenIdsRef.current;
-    const hasNewTraffic = ids.some((id) => !seen.has(id));
+    // Only API-key traffic counts. Studio's own chat goes through these same
+    // endpoints, and this panel is about serving other clients, not about the
+    // request the user is watching stream in front of them.
+    const hasNewTraffic = data.entries.some(
+      (entry) => entry.via_api_key && !seen.has(entry.id),
+    );
     // Re-seed each poll so the set stays bounded by the server's ring buffer.
     seenIdsRef.current = new Set(ids);
     if (!hasNewTraffic) {
diff --git a/studio/frontend/src/features/chat/types/api.ts b/studio/frontend/src/features/chat/types/api.ts
index 6c3e919efe..00f13f87a7 100644
--- a/studio/frontend/src/features/chat/types/api.ts
+++ b/studio/frontend/src/features/chat/types/api.ts
@@ -169,7 +169,10 @@ export interface LoadModelResponse {
   max_context_length?: number | null;
   native_context_length?: number | null;
   supports_reasoning?: boolean;
-  reasoning_style?: "enable_thinking" | "reasoning_effort" | "enable_thinking_effort";
+  reasoning_style?:
+    | "enable_thinking"
+    | "reasoning_effort"
+    | "enable_thinking_effort";
   reasoning_effort_levels?: string[];
   reasoning_always_on?: boolean;
   supports_preserve_thinking?: boolean;
@@ -220,7 +223,10 @@ export interface InferenceStatusResponse {
   } | null;
   requires_trust_remote_code?: boolean;
   supports_reasoning?: boolean;
-  reasoning_style?: "enable_thinking" | "reasoning_effort" | "enable_thinking_effort";
+  reasoning_style?:
+    | "enable_thinking"
+    | "reasoning_effort"
+    | "enable_thinking_effort";
   reasoning_effort_levels?: string[];
   reasoning_always_on?: boolean;
   supports_preserve_thinking?: boolean;
@@ -268,6 +274,9 @@ export interface ApiMonitorEntry {
   model: string;
   prompt?: string;
   reply?: string;
+  // True when the caller used an API key rather than a UI session. The floating
+  // panel keys its auto-open off this so Studio's own chat does not pop it.
+  via_api_key: boolean;
   prompt_preview: string;
   reply_preview: string;
   prompt_truncated: boolean;
@@ -389,7 +398,7 @@ export interface OpenAIChatCompletionsRequest {
     | "xhigh"
     | null;
   preserve_thinking?: boolean | null;
-  thinking?: {type: "disabled" | "enabled";} | null;
+  thinking?: { type: "disabled" | "enabled" } | null;
   enable_tools?: boolean | null;
   enabled_tools?: string[];
   /** Local models + enable_tools only. */