SWE Agent - AI Coding Assistant with Built-in Browser

An AI Coding Agent that helps you write, analyze, and improve code in any programming language. Features its own integrated web browser powered by Julia Browser for complete web interaction capabilities.

Why Use SWE Agent?

Save time writing code - Tell it what you want to build, and it writes the code for you. No more staring at blank files or wrestling with syntax.

Fix bugs faster - It reads your code, finds problems, and suggests fixes. Like having an experienced developer review your work instantly.

Analyze website designs - Upload screenshots and get detailed component breakdowns, color palettes, and layout structures for accurate recreation.

Learn while you code - Watch how it solves problems step by step. You'll pick up new techniques and best practices naturally.

Work in any language - Python, JavaScript, Go, Rust, C++ - it understands them all. Switch between projects without missing a beat.

Stay secure - Automatically scans your code for security issues before you deploy. Catch vulnerabilities early.

Browse the web - Built-in browser can navigate websites, click buttons, fill forms, and extract data automatically.

Deploy to production - Create HTML/CSS/JS applications and deploy them to Netlify with a single command for instant live websites.

Advanced Reasoning Mode - Iterative code improvement with quality targets (8.0/10.0), confidence thresholds (0.85), and reward-based learning over multiple attempts.

Installation

pip install swe-ai-agent

Getting Started

Required environment variables:

# Required: Choose one AI provider
export ANTHROPIC_API_KEY="your-anthropic-api-key-here"
# OR
export OPENROUTER_API_KEY="your-openrouter-api-key-here"

# Optional: For web deployment (only needed if deploying to Netlify)
export NETLIFY_ACCESS_TOKEN="your-netlify-token-here"

Usage Options

# Chat with your AI coding partner
swe-agent --pair

# Run one-time tasks
swe-agent --task "create a todo app with React"

# Work interactively on multiple tasks
swe-agent --interactive

# Use the modern terminal interface
swe-agent --warp

# Enable advanced features with external services
swe-agent --enable-mcp --interactive

# Advanced Reasoning mode with iterative improvement
swe-agent --reasoning --task "create a web scraper" --no-shell-approval

Multi-Provider AI Support ✨

SWE Agent now supports multiple AI providers for maximum flexibility:

Anthropic (Claude)

# Use Anthropic Claude (default)
swe-agent --provider anthropic --task "build a calculator"

# Specify Claude model
swe-agent --provider anthropic --model claude-sonnet-4-20250514 --interactive

OpenRouter (Access 50+ Models)

# Use OpenRouter with Claude
swe-agent --provider openrouter --task "build a website"

# Use OpenRouter with GPT-4o
swe-agent --provider openrouter --model "openai/gpt-4o" --interactive

# Use OpenRouter with Llama
swe-agent --provider openrouter --model "meta-llama/llama-3.1-405b" --pair

# All popular models supported
swe-agent --provider openrouter --model "anthropic/claude-sonnet-4" --warp

OpenRouter Benefits:

🌐 Access to 50+ AI models from different providers
💰 Cost-effective pricing across multiple model options
🔄 Easy model switching without changing code
🚀 Same performance and features across all models
📊 Transparent pricing and usage tracking

New: Web App Deployment ✨

Create and deploy HTML/CSS/JS applications directly to Netlify:

# Create a web application
swe-agent --task "create a portfolio website with HTML and CSS"

# Create AND deploy to Netlify (requires NETLIFY_ACCESS_TOKEN)
swe-agent --task "create a todo app and deploy it to Netlify"

Deployment Features:

🌐 Instant live URLs with SSL certificates
🚀 One-command deployment from creation to production
📱 Mobile-responsive applications by default
🔒 Secure hosting on Netlify's global CDN
🛠️ Smart deployment detection - only deploys when you ask

Advanced Reasoning Mode

The Reasoning Mode implements genuine iterative improvement with:

Quality Target: 8.0/10.0 code quality score
Confidence Target: 0.85 confidence threshold
Max Iterations: 8 improvement attempts
Real-Time UI: Live progress with reward scores and confidence levels
Adaptive Learning: Each iteration learns from previous attempts

# Enable reasoning mode
swe-agent --reasoning --task "optimize this algorithm"

This activates iterative code refinement through multiple attempts, quality assessment, and adaptive improvement until quality targets are met.

Visual Diff Display

See exactly what changes SWE Agent makes to your code with beautiful diff visualization:

# Enable visual diffs (default: enabled)
swe-agent --task "refactor my code" --show-diffs

# Debug mode for detailed operation logs
swe-agent --task "fix the bug" --debug-mode

Visual Features:

🟢 Green borders for file creation
🔵 Blue borders for file edits
🟡 Yellow borders for content replacement
🔴 Red borders for file rewrites
👁️ Vision analysis for website screenshot recreation
Line-by-line diffs with syntax highlighting
Before/after comparisons for all file operations

What It Can Do

Write Code for You

Build complete applications from a simple description
Generate functions and classes that actually work
Create tests for your existing code
Write documentation that makes sense

Deploy Web Applications

Create modern HTML/CSS/JS applications instantly
Deploy to Netlify with professional live URLs
Smart deployment detection (only deploys when explicitly requested)
Mobile-responsive designs with modern web standards

Improve Existing Code

Find and fix bugs before they cause problems
Refactor messy code into clean, readable functions
Add features to existing projects without breaking things
Optimize performance bottlenecks

Keep You Secure

Scan for vulnerabilities in real-time
Suggest security fixes with clear explanations
Check dependencies for known issues
Follow security best practices automatically

Speed Up Your Workflow

Search through large codebases instantly
Navigate complex projects with ease
Track changes and see what was modified
Integrate with git seamlessly

Built-in Web Browser

Navigate websites and click buttons automatically
Fill forms and submit data programmatically
Extract information from web pages
Test web applications interactively
Powered by Julia Browser with 13 browser tools

How It Works

You have four ways to work with SWE Agent:

Pair Programming (--pair) - Like having a coding buddy next to you. Chat naturally about what you want to build, and watch as it writes code, explains decisions, and helps you learn.
Quick Tasks (--task) - Perfect for one-off jobs. "Create a REST API for user management" or "Add authentication to my app" - just describe what you need.
Interactive Sessions (--interactive) - Work on multiple tasks in one session. Great for exploring ideas, iterating on features, and building complex projects step by step.
Terminal Interface (--warp) - A modern coding environment that feels like using advanced developer tools. See your project status, git changes, and AI suggestions all in one place.

Advanced Features

MCP Integration (--enable-mcp)

Enable Model Context Protocol for enhanced capabilities:

Advanced security scanning with Semgrep integration
AI-powered documentation search with DeepWiki
Extended tool ecosystem for specialized tasks
Real-time vulnerability detection and remediation

Reasoning Mode (--reasoning)

Iterative code improvement with quality targets
Confidence-driven decision making
Multi-dimensional reward system
Real-time progress tracking
Adaptive learning and strategy adjustment

Real Benefits

For Beginners: Learn by watching an expert work. See how real applications are built, from planning to deployment.

For Experienced Developers: Focus on architecture and business logic while the AI handles boilerplate, documentation, and testing.

For Teams: Maintain consistent code quality and catch issues before they reach production.

For Everyone: Spend less time debugging and more time building features that matter.

Languages Supported

Works with any programming language - Python, JavaScript, TypeScript, Go, Rust, C++, Java, C#, PHP, Ruby, and more. It adapts to your project's style and conventions automatically.

Safety Features

Secure by default - Scans every change for security issues
Ask before running - Requests permission for potentially dangerous operations
Complete audit trail - Logs everything for transparency
Sandboxed execution - Runs code safely without affecting your system

Getting Help

# See all options
swe-agent --help

# Check if everything is working
swe-agent --status

Start with swe-agent --pair for the best experience. It's like having a senior developer helping you code.

License

MIT License

Stop fighting with code. Start building what you actually want to create.

Built with ❤️ by Harish Santhanalakshmi Ganesan and his Coding AI agents

Browser Integration: Powered by Julia Browser - 43 total tools (30 coding + 13 browser)

Thanks and Credits: Terminal interfaces inspired by Aider and Warp

Release files for swe-ai-agent 2.1.7

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for swe-ai-agent 2.1.7
File	Size	Uploaded
swe_ai_agent-2.1.7.tar.gz	189.5 kB	Aug 15, 2025	Details

Built distribution (wheel)

Table of built distributions (wheels) for swe-ai-agent 2.1.7
File	Interpreter	ABI	Platform	Reset
swe_ai_agent-2.1.7-py3-none-any.whl 213.2 kB Aug 15, 2025	Python 3	none	any	Details

Total release size: 402.7 kB

Release files / swe_ai_agent-2.1.7.tar.gz

Download URL	swe_ai_agent-2.1.7.tar.gz
Size	189.5 kB
Tags	Source
SHA-256 checksum How to use checksums	`0bcd554b925148b63721150b336affdd1cdec0fa381b1a0ee37b21ce7b124bff`
BLAKE2b-256 checksum How to use checksums	`9bae3f280ae4e821526318762f3f7ea5776c1a480613fa9074acd20363790057`
Upload date	Aug 15, 2025
Uploaded using Trusted Publishing? What is trusted publishing?	No
Uploaded via	`twine/6.1.0 CPython/3.11.13`

Release files / swe_ai_agent-2.1.7-py3-none-any.whl

Download URL	swe_ai_agent-2.1.7-py3-none-any.whl
Size	213.2 kB
Tags	Python 3
SHA-256 checksum How to use checksums	`78027c8d5e05c10162dcf06177864771fc14e86cf891c05bf0b264629eca0d9c`
BLAKE2b-256 checksum How to use checksums	`dfee99bde50e2025368a9d66e68b723d8497d506a4467922ea0141d4e789e2ed`
Upload date	Aug 15, 2025
Uploaded using Trusted Publishing? What is trusted publishing?	No
Uploaded via	`twine/6.1.0 CPython/3.11.13`