fix: surface reasoning effort in desktop GUI, delete dead variant knob (fixes #4695) (#4707)

Desktop: add reasoning_effort to DEFAULT_SETTINGS, the settings registry
(Models section) and _llm_overrides(), forwarded through start() only when a
real level is chosen -- "off" stays a no-op so a provider default is never
overridden. Reuses the exact path temperature/top_p already use, no new surface.

CLI: delete InteractiveConfig.variant, a documented "reasoning effort" knob with
zero consumers and no flag to set it.

Co-authored-by: praisonai-triage-agent[bot] <272766704+praisonai-triage-agent[bot]@users.noreply.github.com>
Co-authored-by: MervinPraison <[email protected]>

src/praisonai-code/praisonai_code/cli/interactive/config.py

@@ -64,9 +64,6 @@ class InteractiveConfig:
     # Sharing
     share: bool = False
     
-    # Variant (reasoning effort)
-    variant: Optional[str] = None
-    
     @classmethod
     def from_env(cls) -> "InteractiveConfig":
         """Create config from environment variables."""
@@ -97,7 +94,6 @@ class InteractiveConfig:
             memory=getattr(args, "memory", False),
             files=getattr(args, "file", []) or [],
             share=getattr(args, "share", False),
-            variant=getattr(args, "variant", None),
             autonomy=getattr(args, "autonomy", True),
             no_context=getattr(args, "no_context", False),
         )

src/praisonai-desktop/engine/server.py

@@ -713,6 +713,7 @@ DEFAULT_SETTINGS = {
     "temperature": 0.7,
     "max_tokens": 0,
     "top_p": 1,
+    "reasoning_effort": "off",
     "base_url": "",
     "api_key": "",
     "system_prompt": "",
@@ -1100,6 +1101,11 @@ def _llm_overrides(cfg: dict) -> dict:
         out["max_tokens"] = int(cfg["max_tokens"])
     if cfg.get("top_p") != DEFAULT_SETTINGS["top_p"]:
         out["top_p"] = float(cfg["top_p"])
+    # "off" is the default no-op; anything else is a provider-portable effort
+    # level (minimal|low|medium|high) that Agent maps to each backend's native
+    # knob, exactly as the CLI's --thinking does.
+    if cfg.get("reasoning_effort") not in (None, "", DEFAULT_SETTINGS["reasoning_effort"]):
+        out["reasoning_effort"] = str(cfg["reasoning_effort"])
     _apply_env(cfg)
     return out
 

src/praisonai-desktop/engine/test_chat_stream.py

@@ -44,6 +44,7 @@ class StubAgent:
         self.started = 0
         self.chat_history = []
         self.seen_history = None
+        self.seen_kwargs = None
 
     def _append_to_chat_history(self, message):
         self.chat_history.append(message)
@@ -51,6 +52,7 @@ class StubAgent:
     def start(self, prompt, stream=True, **kwargs):
         self.started += 1
         self.seen_history = list(self.chat_history)
+        self.seen_kwargs = dict(kwargs)
         # Tools "run" before the model's answer would arrive, exactly as the
         # display callback records them during a real turn.
         for tool in self.tools:
@@ -312,6 +314,71 @@ class HistoryAcrossSessions(unittest.TestCase):
                         f"a blank turn was replayed: {agent.seen_history}")
 
 
+class ReasoningEffortReachesTheModel(unittest.TestCase):
+    """The GUI setting must actually arrive at the turn.
+
+    reasoning_effort is a first-class Agent kwarg -- the engine forwards it
+    through start() alongside temperature/top_p, and only when the user chose a
+    real level. "off" is the default no-op and must not be sent, or it would
+    override a provider default the user never touched.
+    """
+
+    @classmethod
+    def setUpClass(cls):
+        cls.server = ThreadingHTTPServer(("127.0.0.1", 0), server.Handler)
+        cls.port = cls.server.server_address[1]
+        cls.thread = threading.Thread(target=cls.server.serve_forever, daemon=True)
+        cls.thread.start()
+
+    @classmethod
+    def tearDownClass(cls):
+        cls.server.shutdown()
+        cls.server.server_close()
+
+    def setUp(self):
+        server._agents.clear()
+        server._tool_queue().clear()
+        self._orig = server.load_settings
+
+    def tearDown(self):
+        server.load_settings = self._orig
+
+    def _install(self):
+        agent = StubAgent(chunks=["ok"])
+        server._agents["t"] = agent
+        server._agents["t\x00notools"] = agent
+        return agent
+
+    def _chat(self):
+        body = json.dumps({"prompt": "hi", "chat_id": "t", "session": "t"}).encode()
+        req = urllib.request.Request(
+            f"http://127.0.0.1:{self.port}/chat", data=body, method="POST",
+            headers={"Content-Type": "application/json"})
+        with urllib.request.urlopen(req, timeout=30) as response:
+            response.read()
+
+    def test_a_chosen_level_is_forwarded_to_start(self):
+        server.load_settings = lambda: dict(self._orig(), reasoning_effort="high")
+        agent = self._install()
+        self._chat()
+        self.assertEqual(agent.seen_kwargs.get("reasoning_effort"), "high",
+                         "the GUI reasoning_effort never reached the turn")
+
+    def test_off_is_the_default_and_is_not_sent(self):
+        # Forwarding "off" unasked would override a provider default the user
+        # never touched, exactly as sending temperature=0.7 unasked would.
+        agent = self._install()
+        self._chat()
+        self.assertNotIn("reasoning_effort", agent.seen_kwargs,
+                         "the no-op default was forwarded and overrode the provider")
+
+    def test_the_setting_round_trips_through_save(self):
+        # It is a real storage key, not a value dropped by load_settings.
+        server.save_settings({"reasoning_effort": "low"})
+        self.assertEqual(server.load_settings()["reasoning_effort"], "low")
+        server.save_settings({"reasoning_effort": "off"})
+
+
 class FetchUrlDoesNotFollowRedirects(unittest.TestCase):
     """The approval card names one URL; the fetch must not reach another.
 

src/praisonai-desktop/frontend/src/settings-registry.js

@@ -47,6 +47,15 @@ const SETTINGS = [
     control: { kind: "slider", min: 0, max: 1, step: 0.01 },
     default: 1, validate: clampNum(0, 1) },
 
+  { key: "reasoning_effort", section: "model", label: "Reasoning effort",
+    description: "How hard the model thinks before answering. Off leaves the provider default. Maps to each backend's native knob.",
+    keywords: ["thinking", "reasoning", "effort", "budget", "o1", "o3"],
+    control: { kind: "select", options: [
+      { value: "off", label: "Off" }, { value: "minimal", label: "Minimal" },
+      { value: "low", label: "Low" }, { value: "medium", label: "Medium" },
+      { value: "high", label: "High" }]},
+    default: "off" },
+
   { key: "base_url", section: "model", label: "Base URL override",
     description: "Point at a proxy, Azure, or a local server. Blank uses the provider default.",
     keywords: ["endpoint", "proxy", "azure", "ollama", "lm studio"],

src/praisonai-desktop/ui/index.html

@@ -676,6 +676,15 @@ const SETTINGS = [
     control: { kind: "slider", min: 0, max: 1, step: 0.01 },
     default: 1, validate: clampNum(0, 1) },
 
+  { key: "reasoning_effort", section: "model", label: "Reasoning effort",
+    description: "How hard the model thinks before answering. Off leaves the provider default. Maps to each backend's native knob.",
+    keywords: ["thinking", "reasoning", "effort", "budget", "o1", "o3"],
+    control: { kind: "select", options: [
+      { value: "off", label: "Off" }, { value: "minimal", label: "Minimal" },
+      { value: "low", label: "Low" }, { value: "medium", label: "Medium" },
+      { value: "high", label: "High" }]},
+    default: "off" },
+
   { key: "base_url", section: "model", label: "Base URL override",
     description: "Point at a proxy, Azure, or a local server. Blank uses the provider default.",
     keywords: ["endpoint", "proxy", "azure", "ollama", "lm studio"],

src/praisonai-desktop/ui/settings-registry.js

@@ -47,6 +47,15 @@ const SETTINGS = [
     control: { kind: "slider", min: 0, max: 1, step: 0.01 },
     default: 1, validate: clampNum(0, 1) },
 
+  { key: "reasoning_effort", section: "model", label: "Reasoning effort",
+    description: "How hard the model thinks before answering. Off leaves the provider default. Maps to each backend's native knob.",
+    keywords: ["thinking", "reasoning", "effort", "budget", "o1", "o3"],
+    control: { kind: "select", options: [
+      { value: "off", label: "Off" }, { value: "minimal", label: "Minimal" },
+      { value: "low", label: "Low" }, { value: "medium", label: "Medium" },
+      { value: "high", label: "High" }]},
+    default: "off" },
+
   { key: "base_url", section: "model", label: "Base URL override",
     description: "Point at a proxy, Azure, or a local server. Blank uses the provider default.",
     keywords: ["endpoint", "proxy", "azure", "ollama", "lm studio"],