Add reasoning command to control reasoning_effort per guild

assistant/assistant.py

@@ -49,6 +49,7 @@ class Assistant(commands.Cog):
             system_prompt="You are a helpful assistant.",
             question_mode=False,
             max_tokens=None,
+            reasoning_effort=None,
         )
         self.config.register_global(api_key=None)
         self.history: dict[int, list[dict]] = {}
@@ -134,6 +135,7 @@ class Assistant(commands.Cog):
         temperature = await self.config.guild(guild).temperature()
         system_prompt = await self.config.guild(guild).system_prompt()
         max_tokens = await self.config.guild(guild).max_tokens()
+        reasoning_effort = await self.config.guild(guild).reasoning_effort()
 
         key = (guild.id, channel_id)
         if key not in self.history:
@@ -147,12 +149,14 @@ class Assistant(commands.Cog):
             "model": model,
             "messages": messages,
             "temperature": temperature,
-            "reasoning_effort": "none",
         }
 
         if max_tokens:
             payload["max_tokens"] = max_tokens
 
+        if reasoning_effort:
+            payload["reasoning_effort"] = reasoning_effort
+
         try:
             async with self.session.post(
                 f"{self.BASE_URL}/chat/completions",
@@ -353,6 +357,23 @@ class Assistant(commands.Cog):
             await self.config.guild(ctx.guild).max_tokens.set(tokens)
             await ctx.send(f"Max response length set to **{tokens}** tokens.")
 
+    @assistant.command(name="reasoning")
+    async def assistant_reasoning(self, ctx: commands.Context, level: str = None):
+        """Set reasoning effort level. Omit to clear (model default).
+
+        Levels: none, minimal, low, medium, high
+        Not all models support all levels. Use 'none' to disable thinking.
+        """
+        valid_levels = ("none", "minimal", "low", "medium", "high")
+        if level is None:
+            await self.config.guild(ctx.guild).reasoning_effort.set(None)
+            await ctx.send("Reasoning effort cleared (model default — thinking enabled).")
+        elif level.lower() not in valid_levels:
+            await ctx.send(f"Invalid level. Choose from: {', '.join(valid_levels)}")
+        else:
+            await self.config.guild(ctx.guild).reasoning_effort.set(level.lower())
+            await ctx.send(f"Reasoning effort set to **{level.lower()}**.")
+
     @assistant.command(name="system")
     async def assistant_system(self, ctx: commands.Context, *, prompt: str):
         """Set the system prompt."""
@@ -385,6 +406,7 @@ class Assistant(commands.Cog):
         sys_prompt = await cfg.system_prompt()
         q_mode = await cfg.question_mode()
         max_tokens = await cfg.max_tokens()
+        reasoning_effort = await cfg.reasoning_effort()
         has_key = await self.config.api_key() is not None
 
         channel_str = f"<#{chat_channel}>" if chat_channel else "None"
@@ -397,6 +419,7 @@ class Assistant(commands.Cog):
         embed.add_field(name="Temperature", value=str(temp), inline=True)
         embed.add_field(name="Question Mode", value=str(q_mode), inline=True)
         embed.add_field(name="Max Tokens", value=str(max_tokens) if max_tokens else "Unlimited", inline=True)
+        embed.add_field(name="Reasoning", value=reasoning_effort if reasoning_effort else "Default (on)", inline=True)
         embed.add_field(name="System Prompt", value=sys_prompt[:200], inline=False)
 
         await ctx.send(embed=embed)