Run OpenAI Codex Locally for FREE with Ollama

Key Concepts

• Ollama: A tool designed to run Large Language Models (LLMs) locally on a user's machine, ensuring data privacy and eliminating subscription costs.
• Codeex: An AI-powered coding assistant/editor that allows for visual website editing and code generation.
• Gemma 2 (Gemma 4): A family of open-weights models from Google used here as the local engine for Codeex.
• Context Window: The amount of text (tokens) the model can consider at once; configured here to 64,000 tokens for optimal performance.
• CLI (Command Line Interface): The terminal-based method for interacting with Ollama and Codeex.

---

1. Setting Up Ollama and Local Models

The process begins by installing Ollama to enable local LLM execution.

• Installation: Users must run the provided installation command in their terminal to set up the Ollama environment.
• Model Selection: The video utilizes Google’s Gemma 2 (referred to as Gemma 4 in the transcript). To download and run the model, the command ollama run gemma2 is used.
• Hardware Considerations: The demonstration uses a Mac Studio with 32GB of RAM, highlighting that performance is dependent on local hardware specifications. Larger models require more memory and processing power.

2. Integrating Codeex with Ollama

Once Ollama is running, Codeex can be configured to use local models instead of cloud-based, paid alternatives.

• Codeex CLI Setup: Install the Codeex CLI via the terminal.
• Launching: Use the command ollama launch codeex. The system will prompt the user to select a model; choosing the locally downloaded Gemma 2 ensures data remains private.
• Codeex App: For a graphical interface, users can download the Codeex app and run ollama launch codeex app to bridge the local Ollama instance with the visual editor.

3. Troubleshooting and Optimization

The video highlights common challenges when using smaller local models for complex tasks:

• Model Capability: Smaller versions of models (e.g., specific "2B" variants) may struggle with advanced tasks like code refactoring or file manipulation.
• Switching Models: If a task fails, users can switch models within the CLI using the /model [model_name] command to utilize a more capable version (e.g., the default Gemma 2).
• Context Window Configuration: To handle larger codebases, the context window must be adjusted. Users should navigate to Ollama Settings and set the context length to 64,000 tokens. This is critical for the AI to "remember" and process larger files effectively.

4. Practical Applications

• Visual Editing: Codeex allows users to select elements on a website and edit them visually, with the AI generating the corresponding code.
• Code Conversion: The tool can be used to refactor code, such as converting a script (app.py) into a class-based structure.
• Data Privacy: Because the entire stack runs locally, sensitive code and data never leave the user's computer, providing a secure alternative to cloud-based AI coding assistants.

5. Synthesis and Conclusion

Running Codeex with Ollama provides a powerful, free, and private alternative to subscription-based AI coding tools. The workflow involves:

1. Installing Ollama and downloading a capable model (Gemma 2).
2. Installing the Codeex CLI or App.
3. Configuring the context window (64k tokens) to ensure the model can handle substantial code files.
4. Managing model selection via the terminal to balance performance with task complexity.

Key Takeaway: While local models offer privacy and cost savings, success depends on matching the model size to the hardware capabilities and ensuring the context window is properly configured for the specific coding task at hand.