Add --reasoning_budget option

core/src/browser/extensions/engines/AIEngine.ts

@@ -126,6 +126,7 @@ export interface loadOptions {
   rope_scale?: number
   rope_freq_base?: number
   rope_freq_scale?: number
+  reasoning_budget?: number
 }
 
 export interface sessionInfo {

extensions/llamacpp-extension/settings.json

@@ -447,5 +447,18 @@
       "placeholder": "path/to/schema.json",
       "type": "text"
     }
+  },
+  {
+    "key": "reasoning_budget",
+    "title": "controls the amount of thinking allowed; currently only one of: -1 for unrestricted thinking budget, or 0 to disable thinking (default: -1)",
+    "description": "Mirostat target entropy (tau).",
+    "controllerType": "input",
+    "controllerProps": {
+        "value": 0,
+        "options": [
+           { "value": -1, "name": "unrestricted thinking budget" },
+           { "value": 0, "name": "disable thinking" }
+     ]
+    }
   }
 ]

extensions/llamacpp-extension/src/index.ts

@@ -347,6 +347,9 @@ export default class llamacpp_extension extends AIEngine {
     if (opts.rope_freq_scale !== undefined) {
       args.push('--rope-freq-scale', String(opts.rope_freq_scale))
     }
+    if (opts.reasoning_budget !== undefined) {
+        args.push('--reasoning-budget', String(opts.reasoning_budget))
+    }
     console.log('Calling Tauri command llama_load with args:', args)
 
     try {