Merge branch 'feature/llm-extra-body-config'

Allows passing arbitrary parameters (e.g., enable_thinking, reasoning_effort)
to the LLM API request body via config.yaml, solving reasoning-only response
loops with models like Qwen 3.x without requiring code changes per model.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

app/models/config.py

@@ -2,6 +2,7 @@
 
 import os
 from pathlib import Path
+from typing import Any
 
 import yaml
 from pydantic import BaseModel, Field, model_validator
@@ -19,6 +20,10 @@ class LLMConfig(BaseModel):
     max_retries: int = Field(default=3, description="Max retry attempts on transient errors")
     retry_backoff_base: float = Field(default=1.0, description="Base seconds for exponential backoff")
     retry_backoff_max: float = Field(default=30.0, description="Maximum backoff seconds")
+    extra_body: dict[str, Any] = Field(
+        default_factory=dict,
+        description="Extra parameters merged into the API request body (model-specific)",
+    )
 
 
 class AgentConfig(BaseModel):

app/services/llm.py

@@ -151,6 +151,10 @@ class LLMClient:
         if tools:
             payload["tools"] = tools
 
+        # Merge model-specific extra parameters (e.g., enable_thinking, reasoning_effort)
+        if self._config.extra_body:
+            payload.update(self._config.extra_body)
+
         try:
             async with self._client.stream(
                 "POST", self._config.api_path, json=payload

config/config.yaml

@@ -10,6 +10,13 @@ llm:
   max_retries: 3
   retry_backoff_base: 1.0
   retry_backoff_max: 30.0
+  # Extra parameters merged into the API request body (model-specific).
+  # Examples:
+  #   Qwen 3.x:  enable_thinking: false
+  #   DeepSeek:   enable_thinking: false
+  #   OpenAI:     reasoning_effort: "low"
+  extra_body:
+    enable_thinking: false
 
 agent:
   max_iterations: 25