We analyzed thousands of developer reviews and completely redesigned the UX/UI so that AI acts as a tool, not a source of stress.

• Smart Loop Breaker: Forget about an AI that repeatedly breaks tests and burns through your tokens. If the agent fails to fix an error 3 times in a row, the system automatically pauses the process, shows a visual Diff, and asks for your help: "Looks like we're stuck. Here is the problem, can you help guide me?"
• Visual Context Manager & 🔒 File Freezing: Complete control over what the AI sees and touches. A convenient UI block shows exactly which files the agent is currently holding in its context. Click "Freeze" (the lock icon) on any file or folder — and the AI won't be able to make a single edit there, even if it thinks it needs to.
• Live Budget Dashboard: No surprises at the end of the month. A bright, compact widget in the status bar shows exactly how many cents/dollars the current prompt or session has cost. Set a Hard Limit (e.g., "stop the agent if the step costs more than $0.50").
• Interactive Onboarding & Architecture Templates: No need to write massive prompts for the Architect. Choose a ready-made preset (e.g., "React+Node CRUD App" or "Refactor using Clean Architecture"), and it will get straight to work.
• Clean UI (Separated AI "Thoughts"): The chat interface is no longer overloaded with walls of text from the internal dialogues of agents. All arguments between the Architect and the Coder are hidden in a dedicated Logs tab, leaving you with just a concise summary and the code Diff.
• Zero-Lag Engine: We rewrote the semantic indexing engine. Kilocode 2.0 is lightning-fast and won't cause UI lags or crash your IDE, even on massive monoliths.

Kilocode doesn't try to solve everything with a single prompt. You have a team of specialists at your disposal:

• Architect: Plans project structure, patterns, and changes before writing code. Perfect for starting a new feature.
• Code (Coder): Writes the actual implementation based on the Architect's plan.
• Debug (Debugger): Hunts for bugs, analyzes logs, reads stack traces, and proposes fixes.
• Ask (Analyst): A safe QA mode for exploring complex codebases without the risk of accidentally modifying files.
• Orchestrator: Manages other agents to execute complex, multi-step tasks.

• BYOK (Bring Your Own Key): Supports 400+ models via OpenRouter integration and our own Kilo Gateway. Use top-tier closed models (Claude 3.7 Sonnet, GPT-4.5/5, Gemini) or free open-source ones (like DeepSeek-V3 or Qwen) to save money.
• Auto Model: Let the system switch models depending on the task to balance cost and quality. Complex planning tasks go to Claude Opus, while routine boilerplate generation goes to faster, cheaper models.

• Memory Bank (Persistent Context): Start a task on your laptop (or phone) and continue on your work PC. Kilocode saves session history, context, and variables.
• MCP (Model Context Protocol) Ecosystem: Connect custom tools from our marketplace. Give the agent terminal access, browser automation (for visual E2E testing), or plugins to read the latest framework documentation.
• AI Code Review: Automatic PR reviews the moment they are created. The agent analyzes the code, suggests improvements, and catches bugs before they merge into main.

Kilocode 2.0 works as a standalone desktop application that seamlessly integrates with your favorite editors.

For Windows:

1. Go to the Releases page.
2. Download the latest Kilocode_2_x64.exe file.
3. Run the installer and follow the on-screen instructions.

For macOS (Apple Silicon / Intel):

1. Go to the Releases page.
2. Download the Kilocode_2_macOS.dmg file.
3. Open the .dmg file and drag the Kilocode icon into the Applications folder.

After launching, the app will offer to automatically install the extensions for VS Code or JetBrains.

1. Open Kilocode.
2. Connect your API key (OpenRouter, Anthropic, OpenAI) or choose the built-in Kilo Gateway.
3. Set your limits in the Live Budget Dashboard.
4. Press Cmd+K (or Ctrl+K) in your editor, choose an architectural template, or just describe the task in your own words!

• How does Kilocode 2.0 prevent infinite AI loops? Thanks to the Smart Loop Breaker, the AI tracks its own success rate. If it fails to fix the same error 3 times in a row, it stops spending your tokens, pauses, and prompts you to provide manual guidance.
• Will the AI randomly rewrite files I don't want it to touch? No. With the new Visual Context Manager and File Freezing, you can lock specific files or entire directories. The AI is strictly blocked from editing any frozen files.
• How can I control my API spending? The Live Budget Dashboard allows you to track expenses in real-time. You can set a Hard Limit (e.g., $1.00 per task), and the Orchestrator will automatically halt if the budget is reached.
• Do I have to use expensive models like Claude Opus for simple tasks? Not at all. The Auto Model feature acts as a smart router: it assigns complex architectural tasks to top-tier models and simple code generation to faster, cheaper models (like DeepSeek or Qwen) to optimize your costs.

This project is licensed under the MIT License - see the LICENSE file for details.