Vajra - Enterprise AI Coding Assistant

The most comprehensive multi-provider AI coding assistant for VSCode

Vajra brings the power of GPT-5, Claude 4, Qwen3-Coder, and 10+ cutting-edge AI models directly into your development workflow. Built as a true Cursor alternative with enterprise-grade features, local model support, and intelligent model routing.

🚀 Key Features

🧠 Latest AI Models (2025)

• GPT-5 & Codex - OpenAI's most advanced coding models (74.9% SWE-bench accuracy)
• Claude 4 Sonnet/Opus - 1M token context window for entire codebase understanding
• Qwen3-Coder - Alibaba's specialized 480B coding model with autonomous agent training
• DeepSeek-Coder V2.5 - Best value with 338 programming languages
• Mistral Codestral 25.01 - Fastest coding assistant with 86.6% HumanEval
• Gemini 2.5 Pro - Multimodal with code execution and 1M context

⚡ Smart Features

• Intelligent Model Routing - Automatically selects the best model for each task
• Multi-Provider Support - 10+ providers including local Ollama models
• Enterprise Security - Local model deployment for sensitive codebases
• Cost Optimization - Transparent pricing with usage tracking
• Multimodal Input - Image, voice, and code understanding

🎯 Coding Superpowers

• Code Generation - Natural language to working code
• Intelligent Refactoring - Context-aware code improvements
• Bug Detection - AI-powered debugging assistance
• Performance Optimization - Automated code optimization
• Test Generation - Comprehensive unit test creation
• Documentation - Auto-generated comments and docs

📦 Quick Start

1. Install the Extension

# Install from VSCode Marketplace
ext install ashishjsharda.vajra

# Or search "Vajra" in VSCode Extensions

2. Configure Your Preferred Provider

For Cloud Models (Recommended):

{
  "vajra.defaultProvider": "qwen",
  "vajra.qwenApiKey": "your-api-key-here",
  "vajra.autoModelSelection": true
}

For Local Models (Privacy-First):

# Install Ollama
curl -fsSL https://ollama.ai/install.sh | sh

# Pull a coding model
ollama pull qwen2.5-coder:7b

3. Start Coding with AI

• Chat: Open sidebar and ask questions
• Code Selection: Right-click code → "Explain with Vajra"
• Quick Actions: Ctrl+Shift+P → "Vajra: Generate Code"

🔧 Supported Providers & Models

☁️ Cloud Providers

OpenAI GPT-5 Series

// Best overall performance
models: [
  'gpt-5',           // 74.9% SWE-bench, 272K context
  'gpt-5-codex',     // Specialized coding model
  'o1-preview',      // Advanced reasoning
  'gpt-4o'           // Multimodal capabilities
]

Anthropic Claude 4

// Best for reasoning and large context
models: [
  'claude-4-sonnet',    // 1M token context window
  'claude-4-opus',      // Premium reasoning model
  'claude-3.5-sonnet'   // Proven coding performance
]

Qwen3-Coder (Alibaba)

// Best for autonomous coding
models: [
  'qwen3-coder-480b-instruct',  // 480B parameter flagship
  'qwen3-coder-32b-instruct',   // Best performance/cost
  'qwen2.5-coder-14b-instruct'  // Efficient option
]

DeepSeek-Coder V2.5

// Best value proposition
models: [
  'deepseek-coder-v2.5',        // Latest unified model
  'deepseek-coder-v2-instruct'  // 338 programming languages
]

Mistral Codestral

// Fastest coding assistant
models: [
  'codestral-25.01',  // 2x faster generation
  'codestral-22b'     // Proven performance
]

🏠 Local Models (Ollama)

Privacy-First Deployment

# Top coding models for local deployment
ollama pull qwen2.5-coder:32b      # Best overall local model
ollama pull deepseek-coder-v2:16b  # Great performance, MIT license  
ollama pull starcoder2:15b         # Open source specialist
ollama pull codellama:34b          # Meta's coding model

⚙️ Configuration

Basic Setup

{
  "vajra.defaultProvider": "qwen",
  "vajra.defaultModel": "qwen3-coder-32b-instruct",
  "vajra.autoModelSelection": true,
  "vajra.temperature": 0.7,
  "vajra.maxTokens": 4096
}

Enterprise Configuration

{
  "vajra.enableCostTracking": true,
  "vajra.enableMultiModalInput": true,
  "vajra.ollamaEndpoint": "http://localhost:11434",
  "vajra.autoModelSelection": true
}

API Key Setup

Go to Settings > Extensions > Vajra and add your API keys:

• OpenAI: Get API Key
• Anthropic: Get API Key
• Qwen: Get API Key
• DeepSeek: Get API Key
• Mistral: Get API Key
• Gemini: Get API Key

🎨 Usage Examples

Generate Code from Natural Language

// Type: "Create a React component for a todo list"
// Vajra generates:

import React, { useState } from 'react';

interface Todo {
  id: number;
  text: string;
  completed: boolean;
}

const TodoList: React.FC = () => {
  const [todos, setTodos] = useState<Todo[]>([]);
  const [input, setInput] = useState('');

  const addTodo = () => {
    if (input.trim()) {
      setTodos([...todos, { 
        id: Date.now(), 
        text: input, 
        completed: false 
      }]);
      setInput('');
    }
  };

  return (
    <div className="todo-list">
      <input 
        value={input}
        onChange={(e) => setInput(e.target.value)}
        onKeyPress={(e) => e.key === 'Enter' && addTodo()}
        placeholder="Add a todo..."
      />
      <button onClick={addTodo}>Add</button>
      <ul>
        {todos.map(todo => (
          <li key={todo.id} className={todo.completed ? 'completed' : ''}>
            {todo.text}
          </li>
        ))}
      </ul>
    </div>
  );
};

export default TodoList;

Intelligent Code Refactoring

// Before (select this code):
function processUsers(users) {
  let result = [];
  for (let i = 0; i < users.length; i++) {
    if (users[i].age >= 18) {
      result.push({
        name: users[i].name,
        email: users[i].email,
        isAdult: true
      });
    }
  }
  return result;
}

// Right-click → "Refactor with Vajra"
// After:
const processAdultUsers = (users: User[]): ProcessedUser[] => {
  return users
    .filter(user => user.age >= 18)
    .map(user => ({
      name: user.name,
      email: user.email,
      isAdult: true
    }));
};

Smart Debugging

// Buggy code:
function calculateTotal(items) {
  let total = 0;
  for (item of items) {  // Missing 'const'
    total += item.price * item.quantity;
  }
  return total;
}

// Select code → "Debug with Vajra"
// Vajra identifies: "Missing 'const' declaration in for-of loop"
// Provides fix with explanation

🏢 Enterprise Features

Security & Compliance

• Local Model Deployment - Keep sensitive code on-premises
• Audit Trails - Complete usage logging and tracking
• SOC 2 Ready - Enterprise security standards
• Data Privacy - No code storage on external servers with local models

Team Management

• Usage Analytics - Track model usage and costs across teams
• Model Policies - Enforce approved models for different projects
• Budget Controls - Set spending limits and alerts

DevOps Integration

• CI/CD Pipeline - Integrate AI code review in automated workflows
• Git Integration - AI-powered commit message generation
• Code Quality - Automated code review and suggestions

🚀 Performance Benchmarks

Benchmarks as of September 2025

💰 Pricing Comparison

API Costs (per 1M tokens)

Local Models (FREE)

• DeepSeek-Coder V2 - MIT License, commercial use allowed
• StarCoder2 - OpenRAIL License, commercial friendly
• CodeLlama - Llama 2 License, <700M users free

🛠️ Advanced Usage

Custom Model Configuration

// Smart model routing based on task
const config = {
  "vajra.modelRouting": {
    "code": "qwen3-coder-32b-instruct",
    "reasoning": "o1-preview", 
    "chat": "claude-4-sonnet",
    "vision": "gpt-5"
  }
};

Multimodal Capabilities

// Upload screenshot for UI-to-code conversion
// Drag & drop images into chat
// Voice commands: "Explain this function"

Batch Processing

// Process multiple files
// Bulk refactoring across codebase
// Generate tests for entire project

🔌 Extensions & Integrations

Popular Integrations

• GitHub Copilot - Use alongside for different strengths
• ESLint/Prettier - Auto-fix with AI suggestions
• Jest/Vitest - AI-generated test cases
• Docker - AI-powered Dockerfile optimization

Language Support

Full Support: TypeScript, JavaScript, Python, Java, C++, C#, Go, Rust, PHP, Ruby, Swift, Kotlin, Scala, R, SQL, HTML, CSS, Bash, PowerShell

Specialized: React, Vue, Angular, Node.js, Django, Flask, Spring Boot, .NET, AWS CDK, Terraform

📄 License

MIT License - see LICENSE file for details.

🆘 Support

• Email: ashishjsharda@gmail.com

⭐ If Vajra helps your coding workflow, please leave a review!

Built with ❤️ by Ashish Sharda