CodeGPT exists in two forms: an open-source VS Code extension and a commercial platform at codegpt.co. Both provide an IDE-integrated AI coding assistant, but the commercial platform adds multi-IDE support (VS Code and JetBrains), an agent marketplace, MCP server connections, and enterprise features like SOC 2 compliance, self-hosting options, and team management. The core workflow lets developers highlight code and ask for explanations, refactoring suggestions, documentation, or unit test generation through a chat panel embedded in the editor — similar to GitHub Copilot Chat but with the flexibility to switch between any AI provider.
What differentiates CodeGPT from built-in IDE AI features is its model-agnostic architecture. Developers can use GPT-5, Claude, Gemini Pro, Llama, Mistral, or any model available through their own API keys — switching models per task to optimize for speed, accuracy, or cost. The platform indexes your entire project structure to provide context-aware suggestions that respect your architecture and coding patterns. Custom AI agents can be created with specific instructions and connected to your codebase, then shared through the marketplace. MCP server integration extends agent capabilities by connecting to external tools, databases, and APIs.
The BYOK pricing model means code flows directly from the IDE to the AI provider through your own API keys — not through CodeGPT's servers. This architecture addresses the data privacy concerns that prevent many enterprises from adopting cloud-based AI assistants. For teams that need additional control, self-hosted deployment options and offline operation with local models via Ollama ensure zero data leaves the organization. CodeGPT provides a free plan with basic features, and the commercial platform offers team licenses with SSO and deployment consulting for enterprises scaling AI-assisted development across their organizations.