Smart-AI-Bridge

Instructions

Implementation Reference

• smart-ai-bridge-v1.1.0.js:1033-1063 (handler)

  Executes the 'ask' tool by mapping the model parameter to a backend, sending the prompt via the MultiAIRouter, recording usage analytics, and handling errors with fallback logging.

  async handleAsk(args) {
    const { model, prompt, thinking = false, force_backend = false } = args;
  
    const startTime = Date.now();
  
    try {
      const backend = model === 'deepseek3.1' ? 'nvidia_deepseek'
                    : model === 'qwen3' ? 'nvidia_qwen'
                    : model;
  
      const result = await this.router.makeRequest(prompt, backend, { thinking });
  
      await this.usageAnalytics.recordInvocation({
        tool_name: 'ask',
        backend_used: backend,
        processing_time_ms: Date.now() - startTime,
        success: true
      });
  
      return result;
    } catch (error) {
      await this.usageAnalytics.recordInvocation({
        tool_name: 'ask',
        backend_used: model,
        processing_time_ms: Date.now() - startTime,
        success: false,
        error
      });
      throw error;
    }
  }

• smart-ai-bridge-v1.1.0.js:460-489 (registration)

  Core registration of the 'ask' tool definition in the coreToolDefinitions array, specifying name, description, handler, and schema. This object is later added to this.coreTools map.

  {
    name: 'ask',
    description: '🤖 MULTI-AI Direct Query - Ask any backend with BLAZING FAST smart fallback chains! Features automatic Unity detection, dynamic token scaling, and response headers with backend tracking.',
    handler: 'handleAsk',
    schema: {
      type: 'object',
      properties: {
        model: {
          type: 'string',
          enum: ['local', 'gemini', 'deepseek3.1', 'qwen3'],
          description: 'AI backend to query: local (Qwen2.5-Coder-7B-Instruct, 128K+ tokens), gemini (Gemini Enhanced, 32K tokens), deepseek3.1 (NVIDIA DeepSeek V3.1, 8K tokens), qwen3 (NVIDIA Qwen3 Coder 480B, 32K tokens)'
        },
        prompt: {
          type: 'string',
          description: 'Your question or prompt (Unity/complex generations automatically get high token limits)'
        },
        thinking: {
          type: 'boolean',
          default: false,
          description: 'Enable reasoning mode for DeepSeek V3.1 (opt-in)'
        },
        force_backend: {
          type: 'boolean',
          default: false,
          description: 'Force use of specified backend even if unhealthy (bypass smart fallback)'
        }
      },
      required: ['model', 'prompt']
    }
  },

• smart-ai-bridge-v1.1.0.js:464-488 (schema)

  Input schema for the 'ask' tool defining properties like model (enum), prompt (required string), thinking (boolean), and force_backend (boolean).

  schema: {
    type: 'object',
    properties: {
      model: {
        type: 'string',
        enum: ['local', 'gemini', 'deepseek3.1', 'qwen3'],
        description: 'AI backend to query: local (Qwen2.5-Coder-7B-Instruct, 128K+ tokens), gemini (Gemini Enhanced, 32K tokens), deepseek3.1 (NVIDIA DeepSeek V3.1, 8K tokens), qwen3 (NVIDIA Qwen3 Coder 480B, 32K tokens)'
      },
      prompt: {
        type: 'string',
        description: 'Your question or prompt (Unity/complex generations automatically get high token limits)'
      },
      thinking: {
        type: 'boolean',
        default: false,
        description: 'Enable reasoning mode for DeepSeek V3.1 (opt-in)'
      },
      force_backend: {
        type: 'boolean',
        default: false,
        description: 'Force use of specified backend even if unhealthy (bypass smart fallback)'
      }
    },
    required: ['model', 'prompt']
  }

• smart-ai-bridge-v1.1.0.js:583-597 (helper)

  Alias definitions that map 'MKG_generate' and 'deepseek_generate' to the core 'ask' tool, enabling backwards compatibility and alternative names for the tool.

    // MKG aliases
    { alias: 'MKG_analyze', coreTool: 'review' },
    { alias: 'MKG_generate', coreTool: 'ask' },
    { alias: 'MKG_review', coreTool: 'review' },
    { alias: 'MKG_edit', coreTool: 'edit_file' },
    { alias: 'MKG_health', coreTool: 'health' },
  
    // DeepSeek aliases
    { alias: 'deepseek_analyze', coreTool: 'review' },
    { alias: 'deepseek_generate', coreTool: 'ask' },
    { alias: 'deepseek_review', coreTool: 'review' },
    { alias: 'deepseek_edit', coreTool: 'edit_file' },
    { alias: 'deepseek_health', coreTool: 'health' }
  ];