LLMs Setting - Centralized AI Provider & System Configuration Hub

LLMs Setting is the centralized configuration dashboard for all AI-powered tools in the T3Lab Revit ribbon extension. It provides a tabbed WPF interface to manage LLM providers, API keys, local model server URLs, action execution safety modes, project workspace boundaries, RAG knowledge store indices, and custom AI skills.

All configuration changes made in LLMs Setting are instantly synchronized across all AI modules (LLMRouter, T3LabAISettings, UserProfile, and ProjectStore) without requiring a Revit restart.

---

5 Core Configuration Tabs

+-----------------------------------------------------------------------------------+
|                                 LLMs Setting Hub                                  |
| +-------------------------------------------------------------------------------+ |
| | [Models] | [General] | [Projects] | [Knowledge (RAG)] | [Skills]                  | |
| +-------------------------------------------------------------------------------+ |
+-----------------------------------------------------------------------------------+

---

Operational Usage Flow & Configuration Sequence

This diagram shows how configuring settings in the LLMs Setting hub propagates credentials, validates live connections, and synchronizes system singletons across all T3Lab tools.

sequenceDiagram
    autonumber
    actor User as "BIM Manager / User"
    participant SettingUI as "LLMs Setting Dialog"
    participant Router as "LLMRouter Singleton"
    participant Config as "settings.json (Disk)"
    participant ProviderAPI as "Cloud API / Local Server"
    participant Assistant as "T3Lab Assistant Chat"

    User->>SettingUI: Open LLMs Setting (Ribbon or Gear Icon)
    SettingUI->>User: Display Models Tab with Active Provider
    User->>SettingUI: Select Provider (e.g. Anthropic Claude) & Model
    User->>SettingUI: Input API Key (sk-ant-...) or Base Host URL
    User->>SettingUI: Click "Test Connection"
    SettingUI->>ProviderAPI: Ping API endpoint with test payload
    alt API Key / Host Valid
        ProviderAPI-->>SettingUI: 200 OK (Latency: 180ms)
        SettingUI-->>User: Display Green Status "Connected & Ready"
    else API Key Invalid / Host Offline
        ProviderAPI-->>SettingUI: 401 Unauthorized / Connection Error
        SettingUI-->>User: Display Red Status "Authentication Failed"
    end
    User->>SettingUI: Switch to "General" Tab
    User->>SettingUI: Select Execution Mode (Manual Confirmation vs Auto Execution)
    User->>SettingUI: Click "Save & Close"
    SettingUI->>Config: Write encrypted settings to settings.json
    SettingUI->>Router: Reload provider singletons & active models
    Router-->>Assistant: Update status indicator in Assistant Toolbar

---

Detailed Step-by-Step Usage Guides

Flow 1: Setting Up Cloud API Providers (Claude / OpenAI / DeepSeek)

1. Open Settings: Click LLMs Setting on the Support panel ribbon or click the Gear Icon inside T3Lab Assistant.
2. Select Provider: On the Models tab, click your desired provider tab:
   • Anthropic Claude: Recommended model claude-3-5-sonnet.
   • OpenAI GPT: Recommended model gpt-4o.
   • DeepSeek: Recommended model deepseek-chat.
3. Enter Credentials: Paste your secret API key into the text box. The UI automatically masks the key for privacy.
4. Test Live Ping: Click Test Connection. A live background probe verifies your key and displays API round-trip latency.
5. Save: Click Save & Close to apply across all T3Lab AI tools.

---

Flow 2: Setting Up Local Offline Engines (Ollama / LM Studio)

1. Start Local Server: Ensure Ollama or LM Studio is running on your local machine.
   • Default Ollama URL: http://localhost:11434
   • Default LM Studio URL: http://localhost:1234
2. Download Model: Open PowerShell/Terminal and run ollama run qwen2.5-coder:7b.
3. Configure in Revit:
   • Open LLMs Setting > Models tab > select Ollama or LM Studio.
   • Type the exact model tag (e.g. qwen2.5-coder:7b).
4. Test Local Ping: Click Test Connection to confirm port reachability.
5. Memory Sizing Tip: In Windows Environment Variables, set T3LAB_OLLAMA_NUM_CTX_MAX = 16384 to prevent high GPU VRAM usage.

---

Flow 3: Managing Project Workspaces & RAG Knowledge Store

1. Projects Tab: Link active Revit files (doc_key) to designated project folders. Project-specific BIM Execution Plans (BEP) and rules will automatically scoped to the model.
2. Knowledge Tab:
   • Ingest PDF guidelines, Revit modeling standards, and client specs into vector embeddings.
   • Click Rebuild RAG Index to update vector search references.
3. Skills Tab: View active custom skills and add new automation scripts to expand T3Lab Assistant capabilities.

---

System Integration & Shared Singletons

All settings saved in LLMs Setting persist to config/settings.json and are shared immediately by:

• LLMRouter: Handles AI request routing and model completions.
• T3LabAISettings: Stores execution safety flags, language defaults, and provider configurations.
• UserProfile: Manages user credentials and roles.
• ProjectStore: Manages workspace directories and project boundaries.