No, Gemini CLI itself does not support direct fine-tuning with your own data. The CLI tool is designed to provide access to pre-trained Gemini models through Google’s cloud infrastructure, and fine-tuning capabilities are not built into the command-line interface. Fine-tuning requires significant computational resources and specialized infrastructure that the CLI tool doesn’t provide locally. The tool is optimized for providing immediate access to powerful, pre-trained models rather than customizing models with specific datasets.
However, fine-tuning capabilities for Gemini models are available through other Google platforms, specifically through Vertex AI and Google AI Studio. These platforms support supervised fine-tuning where you can customize Gemini models using your own labeled datasets. The fine-tuning process requires uploading datasets to Cloud Storage buckets and configuring tuning jobs through the respective platforms. You can fine-tune models like Gemini 2.5 Flash for specific tasks such as classification, sentiment analysis, entity extraction, or domain-specific text generation. The process involves providing training datasets with input-output pairs that teach the model to perform your specific tasks.
If you need fine-tuned capabilities accessible from the command line, you would need to fine-tune a model through Vertex AI or Google AI Studio first, then potentially access that custom model through the CLI using appropriate API credentials and model specifications. Alternatively, you can achieve similar customization effects through Gemini CLI’s configuration system using GEMINI.md files to provide project-specific instructions, coding standards, and context that guide the model’s behavior for your particular use case. While this isn’t true fine-tuning, it can help tailor the model’s responses to your specific needs and workflows without requiring the computational overhead and technical complexity of actual model fine-tuning.