Vishwa is an AI-powered terminal coding assistant that thinks, reasons, and acts. Named after Vishwakarma, the divine architect, it brings intelligent code assistance right to your terminal and VS Code.
Vishwa is an AI-powered coding assistant with two main components: a terminal agent and VS Code autocomplete functionality. Here's how it works:
Vishwa (named after Vishwakarma, the Hindu god of engineering) operates as a dual-purpose coding assistant:
• ReAct Agent Loop: Uses a reasoning-acting cycle to break down complex tasks
• Multi-LLM Support: Supports Claude, GPT-4, Ollama (local), and Novita providers
• Core Tools: Bash, file operations (read/write/edit), git operations, web search
• Configuration: Managed through ~/.vishwa/.env file with model selection
• Background Service: Runs as a Python service communicating via JSON-RPC over stdio
• Real-time Suggestions: Provides AI-powered code completions as you type
• Caching System: Uses LRU cache with TTL for performance (100 suggestions, 5min TTL)
• Context Awareness: Analyzes surrounding code to provide relevant suggestions
Vishwa combines the power of multiple AI models with a sophisticated tool ecosystem to help you code smarter, not harder.
Think, act, observe, repeat. Vishwa uses the ReAct pattern to iteratively solve complex problems with up to 15 reasoning iterations per task.
Use Claude Sonnet 4.5, GPT-5.1, or open-source models like Kimi K2 and GLM-4.6 via Novita. Switch seamlessly with intelligent fallback chains.
Ultra-fast suggestions with <100ms latency. Context-aware completions that understand your code structure as you type.
File operations, git commands, code analysis, web search, and more. Vishwa has the tools to handle any coding task.
Run entirely local with Ollama models. Your code never leaves your machine when you choose privacy-first mode.
User approval for file changes, surgical string replacements, and git integration for easy rollback. Your code is protected.
A sophisticated agentic loop that thinks before it acts
Describe what you want to accomplish in natural language
AI analyzes your codebase and plans the approach
Specialized tools read, analyze, and modify your code
Approve changes and iterate until satisfied
AI-powered code completions that understand your codebase
<100ms latency with local models, smart caching delivers 50-70% cache hit rates
Analyzes surrounding code, imports, and file structure for relevant suggestions
Ghost text suggestions appear automatically as you type - Tab to accept, Esc to dismiss
Python, JavaScript, TypeScript, Go, Rust, and more - works with any language
from flask import Flask, requestfrom datetime import datetimeapp = Flask(__name__)@app.route('/api/users')def get_users(): return jsonify({'users': [], 'timestamp': datetime.now().isoformat()})Use the best model for your task, or run completely offline
Choose your platform and start coding with AI
git clone https://github.com/srujan375/Vishwa.git && cd Vishwa && ./install.sh
export ANTHROPIC_API_KEY="your-key-here"
vishwa "Explain this codebase"
git clone https://github.com/srujan375/Vishwa.git; cd Vishwa; .\install.ps1
$env:ANTHROPIC_API_KEY="your-key-here"
vishwa "Explain this codebase"
git clone https://github.com/srujan375/Vishwa.git && cd Vishwa
python -m venv .venv && source .venv/bin/activate
pip install -e .
Want to use local models? Install Ollama and run ollama pull gemma3:4b