[Feature Request] Advanced Model UI: Reasoning Levels, Search Toggles & OpenRouter Fetching

[sergioatp]

Hi @glowingjade,

First off, thank you for the incredible work on Smart Composer. It has become an essential tool in my workflow.

I’m opening this discussion to propose a few UI enhancements that would make managing the latest generation of models (specifically Grok 4.1, Claude Opus 4.1, and GPT-5 via OpenRouter) significantly easier.

Currently, to utilize the full capabilities of these models—such as Thinking/Reasoning levels or specific RAG/Search permissions—we have to manually edit the data.json file. This is error-prone and brittle, as settings can be overwritten.

Here is a breakdown of the friction points and a proposed solution.

---

1. The Problem: Manual JSON Editing for Model Capabilities

Right now, if I want to use x-ai/grok-4.1-fast or anthropic/claude-opus-4.1 and control the reasoning depth, I have to manually create "virtual models" in data.json and hardcode the parameters like this:

// I have to manually create this entry to force High Reasoning
{
  "id": "grok-4.1-fast-high",
  "providerId": "openrouter",
  "model": "x-ai/grok-4.1-fast",
  "name": "Grok 4.1 Fast (Max Reason)",
  "capabilities": { 
      "thinking": true, 
      "search": true 
  },
  "extraBody": {
    "reasoning": { "effort": "high" } // Hardcoded parameter
  }
}

If I want to switch to "No Reasoning" for a faster response, I have to duplicate the entry and change the JSON manually.

2. The Solution: "Model Capabilities" UI Menu

It would be fantastic if the Edit Model modal included a "Capabilities" section with the following controls:

• ☑ Enable Thinking Module:
  • Action: Sets "capabilities": { "thinking": true } in the config.
  • Visual: Enables the <think> tag visualization in the chat (critical for models like Claude Opus 4.1 and DeepSeek R1).
• Reasoning Effort (Dropdown):
  • Options: Default | Low | Medium | High
  • Action: Automatically injects the correct parameter into the request body based on the provider standard.
    • For OpenRouter/Anthropic: {"reasoning": {"effort": "high"}}
    • For OpenAI (GPT-5/o1): {"reasoning_effort": "high"}
• ☑ Enable Search (RAG):
  • Action: Sets "capabilities": { "search": true }. Allows us to explicitly enable/disable RAG for specific high-cost models.

3. OpenRouter Enhancements

Since many of us rely on OpenRouter for models like GPT-5 and Grok 4.1, it would be incredibly helpful to have:

• Provider Filtering: A way to filter the model list by Provider in the settings menu (e.g., "Show only OpenRouter models").
• Fetch Models: A button to query the OpenRouter API and populate the list automatically with the correct IDs (e.g., x-ai/grok-4.1-fast), preventing ID typos.

Why this matters

With the release of Grok 4.1 and GPT-5, the distinction between "Reasoning" and "Standard" chat is becoming a standard feature. Having a native UI to toggle these levels would make Smart Composer the most powerful, user-friendly interface for Obsidian users who want to leverage these bleeding-edge capabilities.

Thank you for considering this!

---

[sergioatp]

Workaround: How to Enable "Reasoning Effort" UI for OpenRouter Models (Grok, Gemini, etc.)

I found a way to unlock the Reasoning Effort dropdown and settings menu for OpenRouter models without modifying the plugin code.

The Problem:
The plugin currently only renders the "Gear" icon and "Reasoning Effort" controls if the providerType is strictly openai (or anthropic for thinking). Custom OpenRouter models usually lose these UI controls.

The Solution:
You can "masquerade" OpenRouter as the OpenAI provider in your data.json. By pointing the OpenAI baseUrl to OpenRouter and defining your models as providerType: "openai", the plugin enables all the advanced UI controls, while OpenRouter handles the routing perfectly.

Configuration:
Edit your .obsidian/plugins/smart-composer/data.json:

1. Hijack the OpenAI Provider: Set the baseUrl to OpenRouter and add noStainless to headers.
2. Define Models: Add your Grok/Gemini models but set their providerType to "openai".

{
  "providers": [
    {
      "type": "openai",
      "id": "openai",
      "apiKey": "sk-or-v1-YOUR_OPENROUTER_KEY",
      "baseUrl": "https://openrouter.ai/api/v1",
      "additionalSettings": {
        "noStainless": "true"
      }
    }
    // ... other providers
  ],
  "chatModels": [
    {
      "providerType": "openai", 
      "providerId": "openai",
      "id": "grok-4.1-fast-reasoning",
      "model": "x-ai/grok-4.1-fast",
      "reasoning": {
        "enabled": true,
        "reasoning_effort": "high"
      }
    },
    {
      "providerType": "openai",
      "providerId": "openai",
      "id": "gemini-3-pro-thinking",
      "model": "google/gemini-3-pro-preview",
      "reasoning": {
        "enabled": true,
        "reasoning_effort": "high"
      }
    }
  ]
}

Result:
The plugin now treats these as OpenAI models, displaying the Settings Gear next to them. You can toggle Reasoning on/off and change the Effort (Low/Medium/High) directly from the UI, and it works perfectly with OpenRouter's API standard.

[Lapis0x0]

Maybe you want to try YOLO?Here is the link：https://github.com/Lapis0x0/obsidian-yolo