OpenCode just killed all vibe coding apps (it’s insane)

How to Build Anything with Open Code

Introduction to Open Code

• David Andre introduces Open Code as a rapidly growing coding agent, surpassing Cloth Code in popularity.

• Highlights that Open Code is free, compatible with over 70 AI models, and allows users to choose any LLM provider.

Installation Process

• Instructions for installation via terminal using the curl command; emphasizes the importance of having the latest version (1.16).

• Discusses various providers available through Open Code and mentions its open-source nature, making it more appealing than closed alternatives.

Setting Up API Key

• Guides viewers on creating an account at openrouter.ai to obtain an API key for accessing different models.

• Completes setup by pasting the API key into the terminal, demonstrating how simple it is to get started.

Using Open Code in IDE

• Launches Open Code from an IDE (Cursor), explaining how to set a model using CLI commands.

• Demonstrates creating a 3D endless runner game using Vanilla TypeScript and 3JS; outlines project structure and features.

Project Execution and Features

• Describes how Open Code generates a task list before writing code, showcasing its organized approach.

• Notes the user-friendly interface with real-time updates on tasks completed and token usage during project execution.

Game Development Outcome

• Confirms successful compilation of the game after running npm install commands without errors.

• Concludes with gameplay demonstration; highlights potential for further development like adding levels or camera angles based on initial prompt.

Open Code: The Future of AI Coding Agents

Recent Developments in Open Code

• Discussion on the recent drama involving Enthropic, which restricted third-party apps from using Clawed credentials, impacting Open Code's functionality.

• Enthropic's aggressive move highlights the value of their models, as many companies, including XAI, sought to utilize them for software development.

• Following backlash, Open Code quickly pivoted by partnering with OpenAI and GitHub to allow users to integrate their subscriptions into Open Code.

Features and Performance of Open Code

• Demonstration of building a CRM dashboard for email campaigns using Gemini 3 Pro within Open Code; aims for a seamless one-shot execution.

• Emphasis on the ease of use with Open Code compared to Clawed code; highlights its beginner-friendly interface and additional features like token tracking.

User Experience Comparison

• Comparison between user interfaces: Clawed code is noted for its clean design but lacks some functionalities that enhance user experience in Open Code.

• Highlighting quality-of-life improvements in Open Code such as automatic text copying and detailed status lines that provide real-time information about models being used.

Application Testing Results

• Successful completion of the CRM application without errors; all features function correctly during testing (e.g., adding and deleting contacts).

• Observations on UI aesthetics; while functional, there are suggestions for further customization to improve uniqueness.

Customization and Themes

• Discussion on theme customization options available in Open Code compared to limited choices in Clawed code; flexibility allows users to personalize their coding environment.

• Mention of default themes versus customizable options that reflect personal style or preferences among developers.

Clarification on What is Open Code

• Explanation of what constitutes an AI coding agent like Open Code; operates primarily through CLI but also offers a desktop app for broader accessibility.

• Acknowledgment of confusion surrounding new AI tools emerging weekly; clarifies the specific role and capabilities of Open Code within this landscape.

Open Code vs. Cloud Code: A Detailed Comparison

Overview of Open Code

• Open Code is described as a coding agent rather than a full IDE, distinguishing it from platforms like VS Code and others.

• The speaker has extensive experience with Cloud Code, having spent over 500 hours using it and creating numerous videos about it.

Key Differences Between Open Code and Cloud Code

• Open Code is open source under an MIT license, while Cloud Code is closed source, which raises concerns regarding privacy and security.

• Open Code supports over 70 different AI models (e.g., OpenAI, Gemini), whereas Cloud Code only supports cloud models due to its proprietary nature.

Cost Effectiveness

• Using GitHub Copilot as the backend for Open Code costs $10/month for 500 premium requests, making it more affordable compared to Cloud Code's plans that range from $100 to $200 per month.

• The rapid development of features in Open Code is attributed to its open-source nature, allowing community contributions that enhance the project quickly.

Building Applications with Open Code

• The speaker demonstrates building a web-based drawing editor using GPD 5.2 Codex within Open Code, showcasing flexibility in model selection.

• Users can log into Open Code using their existing subscriptions (e.g., CHBD), which can make the service effectively free if they already pay for those subscriptions.

Prompt Structuring for Efficiency

• Effective prompts are crucial; the speaker emphasizes structuring them clearly to facilitate quick app development without errors.

• The process includes creating necessary files and directories rapidly, highlighting the efficiency of GBD 5.2 Codex when used with Open Code.

Performance Insights

• GBD 5.2 Codex operates significantly faster than other models like Opus or Gemini 3 during application builds.

• Users are encouraged to leverage their existing subscriptions with multiple coding agents instead of incurring high costs with proprietary services like Enthropic’s offerings.

Open Code: Enhancing Development Efficiency

Benefits of Open Code

• The speaker highlights the advantage of using Open Code, noting that it allows for a single interface across all models, simplifying user experience compared to various CLI tools like Gemini CLI and Cloud Code.

• Emphasizes the ease of starting with an existing codebase by using the command slash init, which initializes the agents.mmd file to analyze and learn from the entire codebase.

Understanding agents.mmd

• The agents.mmd file serves as a convention similar to a README, originally developed by OpenAI and now under the Linux Foundation, adopted by multiple tools including Open Code.

• Various coding tools such as GitHub Copilot and Warp utilize the agents.mmd file, indicating its importance in enhancing agent performance within a codebase.

Practical Application of agents.mmd

• Demonstrates how to launch an Open Code agent and use slash init to create an agents.mmd file that includes build commands, style guidelines, and other essential information for coding agents.

• Highlights that if an existing agents.mmd is found, it can be improved upon rather than created anew, showcasing efficiency in utilizing prior work.

Performance Insights

• Discusses how having an agents.mmd file significantly boosts AI agent performance; beginners often miss out on this advantage due to lack of knowledge about effective prompting.

• Notes that while Codex may take longer due to its thoroughness in processing changes, this depth can be beneficial when addressing complex bugs.

Contextual Importance for AI Agents

• Stresses that context is crucial for AI performance; having detailed information in the agents.md enhances future agent interactions with the codebase.

• Observes that creating an agents.md took 2 minutes 48 seconds—considered slow but results in well-formatted output essential for efficient development.

User Experience with Open Code

• After generating the necessary files, demonstrates running local development servers via npm commands and showcases features like screenshot capabilities within a simplified drawing tool.

• Compares functionalities between TL Draw and Photoshop; notes TL Draw's additional features despite being less polished overall.

CLI Experience with Open Code

• Concludes with praise for Open Code's terminal UI experience which includes automatic prompt summarization and task management features—potentially superior to competing tools.

Understanding Open Code and Its Advantages

Modes of Operation in Open Code

• Open code features two modes: plan mode and build mode, which can be toggled using the tab key. This functionality enhances coding efficiency, especially for VIP coding.

• The CLI experience is notably superior due to its development by the Neoim team, who possess a deep understanding of terminal interfaces and user expectations.

Support for Agents.md

• A significant advantage of open code is its support for agents.mmd, a convention initiated by OpenAI, unlike cloth code which lacks this feature.

• For global configurations, files must reside at the root level; project-specific settings require an hgn.MD file created with the /init command.

Building Encrypted File Manager Software

• The speaker plans to develop an encrypted file manager software to safeguard personal data amidst increasing centralization and data privacy concerns.

• Using GLM 4.7 through open router allows seamless session management without repeated authentication, enhancing user convenience.

Importance of Context in AI Development

• In plan mode, the AI prompts users with questions that help clarify context—an essential factor for achieving optimal results when building applications quickly.

• The speaker emphasizes providing adequate context to avoid common pitfalls that hinder productivity in AI-assisted development.

Prototyping and Iteration Process

• The prototype is described as "quick and dirty," indicating a focus on rapid development rather than perfection at this stage.

• If issues arise during execution (e.g., model performance), users can interrupt processes easily by pressing escape twice—a more efficient method compared to closed code.

Evaluating Model Performance

Challenges with GLM 4.7

• There are uncertainties regarding whether errors stem from model limitations or external factors like system inactivity during breaks.

• While GLM 4.7 is fully open source, it may not perform as well as other models like Opus 4.5 or GPT 5.2 CEX; however, its accessibility remains a strong point.

Value of Coding Agents

• Effective coding agents leverage not just the model but also provide necessary tools and prompts within their environment to enhance performance significantly.

Future Prospects in Software Development

Comparing Coding Platforms

• The future landscape will reveal how platforms like cloud core and open code evolve over time concerning their capabilities and user experiences.

Enhancements in User Experience

• Users benefit from improved quality-of-life features such as remembering recent models used, streamlining workflows significantly compared to previous systems like wind surf.

Privacy and Open Source in AI Development

The Importance of Privacy-First Approaches

• Emphasizes the significance of a privacy-first approach in AI, highlighting that certain principles cannot be altered.

• Mentions "Agent Zero" as a preferred choice due to its local run capability, open-source nature, and commitment to privacy and security.

• Contrasts with Enthropic, which is closed source and collects user data for training purposes.

Open Source vs. Closed Source

• Discusses the inherent differences between open source (like Agent Zero) and closed source systems (like Enthropic), stressing that foundational principles remain unchanged despite feature updates.

• Predicts that the value of privacy-focused tools will increase over time as users become more aware of data collection practices.

User Interaction with AI Tools

• Describes an improved method for inputting queries into AI tools, favoring multi-selection over manual typing for efficiency.

• Highlights a new predictive feature in Claw code that anticipates user prompts, enhancing workflow by saving time during coding tasks.

Risks Associated with AI Commands

• Expresses concern about sending data to Enthropic while using their services; emphasizes the importance of being cautious with commands generated by AI.

• Warns against executing risky recursive deletion commands without verification, advocating for thorough checks before running such commands.

Enhancements in User Experience

• Notes improvements in how open code interacts with system files, allowing users to access folders directly from terminal commands—an unexpected but beneficial feature.

• Concludes on the effectiveness of Opus compared to other models like GLM, indicating progress in handling technical inconsistencies within file management tasks.

Encrypted File CLI Tool Overview

Introduction to the Tool

• The tool updates to-do lists and demonstrates resilience by self-fixing errors during its build process, showcasing the robustness of open-source code.

• It features a command-line interface (CLI) for securely storing and retrieving files using AES 256 encryption.

Testing the Encryption Process

• The speaker plans to test the encryption with a large file (500 MB MP4), questioning how long it will take for encryption or decryption.

• Demonstrates ease of use by copying commands directly into the terminal, highlighting user-friendly design elements like password confirmation.

User Experience During Encryption

• Discusses unlocking the vault and adding files, noting potential issues with file names containing spaces.

• Suggests that feedback on encryption progress in the CLI would enhance user experience; notes that encryption took about 20 seconds.

Cleanup and Usability Concerns

• Requests guidance on clearing test files post-testing to avoid cluttering storage space.

• Critiques scrolling sensitivity in the interface, suggesting improvements for smoother navigation compared to other tools.

Open Code Features and Accessibility

Native Desktop Application

• Emphasizes that Open Code is not limited to CLI but also offers a native desktop application, which is more beginner-friendly.

Integration with Development Tools

• Mentions an extension available for VS Code users, making installation straightforward for those who prefer graphical interfaces over terminals.

Comparison of Interfaces

• Highlights that while both interfaces are effective, the desktop app is particularly suitable for beginners due to its simplicity. The CLI version is noted as having one of the best UIs among similar tools.

Open Code Zen: A Cost-Efficient Solution

Subscription Model Explained

• Introduces Open Code Zen as an optional service providing cost-effective access without markup on model usage fees.

Usage-Based Pricing Benefits

• Describes how users only pay based on actual usage rather than fixed monthly fees, contrasting this with traditional subscription models where limits may not be fully utilized.

Available Models and Offerings

• Lists various models available through Open Code Zen including free options like Big Pickle and Gro code fast one aimed at attracting new users.

Grock 4.1 Fast: A Complex Tool for AI Coding

Overview of Grock 4.1 Fast

• The speaker introduces Grock 4.1 fast, describing it as a complex tool with advanced features aimed at enhancing coding efficiency.

• There is uncertainty about which Grock model excels in coding, but the speaker opts to use Grock 4.1 for its speed and capabilities.

Comparison with Codex

• The speaker highlights a disadvantage of using the GPD502 model compared to Codex, specifically regarding the inability to set reasoning effort levels.

• Codex allows users to select reasoning effort (low, medium, high), while Grock defaults to high or extra high, which may be excessive for most tasks.

Advantages and Disadvantages

• An objective assessment reveals that while open code has advantages like being open-source and community-driven, it also has limitations compared to proprietary tools.

• The speaker notes that closed-source companies often have an edge due to exclusive features not available through APIs.

Performance Insights

• Despite claims of performance benchmarks favoring Grock 4.1 fast, the speaker argues that it does not perform as well as Opus or Gemini models in practical applications.

• Users are reportedly more inclined towards Opus 4.5 or Gemini 3 for coding tasks rather than relying on Grock 4.1 fast.

Importance of Deep Research

• The discussion shifts towards utilizing deep research tools like Plexity D for better API usage insights when developing applications.

• Emphasizing the need for thorough research before coding can significantly enhance problem-solving capabilities beyond just programming challenges.

Common Misconceptions About AI Tools

• Many users fail to leverage powerful reasoning models effectively; this oversight limits their ability to solve various life problems creatively.

• The speaker critiques those who believe they are proficient with AI tools without fully understanding how to utilize them effectively.

Final Thoughts on Model Performance

• A strong comparison indicates that Opus performs significantly better than Grock 4.1 fast in real-world scenarios; users should rely on personal experience rather than benchmarks alone.

• The conclusion emphasizes testing models personally instead of solely depending on reported performance metrics from others.

Project Setup and Error Handling

Initial Project Configuration

• The project structure is introduced, emphasizing the need to copy the .env.local file for configuration.

• A reminder is given about the importance of keeping API keys confidential, treating them like passwords.

Encountering Errors

• Upon running the application, multiple errors are encountered; a screenshot of these errors is taken for reference.

• The speaker discusses troubleshooting by switching to "plan mode" to analyze issues without executing changes.

Simplifying Development Modes

• The concept of "open code" is highlighted as a way to simplify development by focusing on two modes: plan and build.

• The speaker criticizes overly complex coding tools with numerous modes, advocating for a straightforward approach where planning precedes building.

Debugging and UI Improvements

Switching Between Modes

• An error occurs due to not switching back to build mode after planning; this highlights the importance of managing development states effectively.

• Features such as message history and easy navigation within the UI are praised for enhancing user experience during debugging.

Addressing Model Issues

• The speaker notes discrepancies in model names used in prompts, indicating potential issues with outdated contexts or incorrect configurations.

• A new spec.md file is created to clarify model specifications and prevent confusion in future interactions with AI models.

User Interface Feedback and Functionality Testing

Evaluating Frontend Design

• Criticism is directed at the application's frontend design, comparing it unfavorably to modern standards; suggestions for improvement are made.

• After reloading, some improvements in UI design are noted, leading to successful image uploads for processing.

Image Processing Challenges

• The application attempts an API call for image enhancement but encounters delays and errors (500 error), indicating backend issues during processing.

Identifying Core Issues

Troubleshooting Output Generation

• A specific issue arises when selecting variations generated by Opus 4.5; while they display correctly, outputs from Narabara Pro do not appear as expected.

Documentation Reference

• New documentation files are created to ensure adherence to coding standards and facilitate better understanding among team members regarding implementation steps.

Open Code: A Powerful Tool for Developers

Importance of Documentation in Development

• Emphasizes the necessity of creating a markdown file within the codebase for essential information, allowing multiple tags for better organization.

Advancements in AI and Image Generation

• Discusses the rapid evolution of AI tools since 2022, highlighting improvements in generating variations and enhancing image quality, particularly lighting issues.

Debugging Challenges with AI Tools

• Describes an issue where the canvas resets unexpectedly during image generation attempts, indicating potential bugs or limitations in current AI implementations.

Application Versatility of AI Tools

• Notes that AI can be utilized across various sectors such as e-commerce and dating apps, stressing the importance of persistence and continuous learning to achieve impressive results.

Learning Curve and Technical Skills

• Points out that while simple applications can be created quickly using AI coding tools, unique software requires deeper technical knowledge and skills to develop effectively.

File Format Compatibility Issues

• Raises questions about file format compatibility (e.g., WEBP vs. PNG), suggesting that converting files may simplify processes when facing technical challenges.

Performance Evaluation of Open Code

• Reflects on a mixed success rate (4 wins, 1 loss) when building complex applications with Open Code, acknowledging its capabilities but also recognizing areas needing improvement.

Comparison with Other Coding Models

• Compares Open Code's performance against other models like Cloth Code, noting its advantages in utilizing cloud models effectively due to its optimized environment.

Modes of Operation: Plan Mode vs. Build Mode

• Explains the distinction between Plan Mode (read-only analysis without changes to files) and Build Mode (allows writing/deleting files), emphasizing safety in development practices.

User Experience Enhancements in Open Code

• Highlights user-friendly features such as automatic summarization of headings and integrated to-do lists that improve navigation and usability within Open Code's interface.

The Rise of Open Code as a Competitor

• Concludes by noting the growing popularity of Open Code among developers due to its powerful features, open-source nature, and cost-effectiveness for existing subscribers.