gemini-code 0.1.34

An AI coding assistant CLI using Google's Gemini models with function calling.

Gemini Code

A powerful AI coding assistant for your terminal, powered by Gemini 2.5 Pro with support for other LLM models.

Features

• Interactive chat sessions in your terminal
• Multiple model support (Gemini 2.5 Pro, Gemini 1.5 Pro, and more)
• Intelligent context management with auto-compaction warnings and the /compact command
• Markdown rendering in the terminal
• Automatic tool usage by the assistant:
  • File operations (view, edit, list, grep, glob)
  • System commands (bash)
  • Web content fetching

Installation

# Clone the repository
git clone https://github.com/raizamartin/gemini-code.git
cd gemini-code

# Install the package
pip install -e .

Setup

Before using Gemini CLI, you need to set up your API keys:

# Set up Google API key for Gemini models
gemini setup YOUR_GOOGLE_API_KEY

Usage

# Start an interactive session with the default model
gemini

# Start a session with a specific model
gemini --model models/gemini-2.5-pro-exp-03-25

# Set default model
gemini set-default-model models/gemini-2.5-pro-exp-03-25

# List all available models
gemini list-models

Interactive Commands

During an interactive session, you can use these commands:

• /exit - Exit the chat session
• /help - Display help information
• /compact - Summarize the conversation to reduce token usage

How It Works

Tool Usage

Unlike direct command-line tools, the Gemini CLI's tools are used automatically by the assistant to help answer your questions. For example:

1. You ask: "What files are in the current directory?"
2. The assistant uses the ls tool behind the scenes
3. The assistant provides you with a formatted response

This approach makes the interaction more natural and similar to how Claude Code works.

Context Management

Gemini CLI intelligently manages the conversation context:

1. Warning Threshold (80%): When you reach 80% of the token limit, you'll see a warning panel suggesting to use /compact
2. Auto-Compact Prompt (95%): At 95% of the limit, the CLI will ask if you want to automatically compact the conversation
3. Manual Compaction: You can use /compact at any time to summarize the conversation and reduce token usage

The summarization process preserves important context while significantly reducing token count, allowing for virtually unlimited conversation length.

Development

This project is under active development. More models and features will be added soon!

Recent Changes in v0.1.21

• Implemented native Gemini function calling for much more reliable tool usage
• Rewrote the tool execution system to use Gemini's built-in function calling capability
• Enhanced the edit tool to better handle file creation and content updating
• Updated system prompt to encourage function calls instead of text-based tool usage
• Fixed issues with Gemini not actively creating or modifying files
• Simplified the BaseTool interface to support both legacy and function call modes

Recent Changes in v0.1.20

• Fixed error with Flask version check in example code
• Improved error handling in system prompt example code

Recent Changes in v0.1.19

• Improved system prompt to encourage more active tool usage
• Added thinking/planning phase to help Gemini reason about solutions
• Enhanced response format to prioritize creating and modifying files over printing code
• Filtered out thinking stages from final output to keep responses clean
• Made Gemini more proactive as a coding partner, not just an advisor

Recent Changes in v0.1.18

• Updated default model to Gemini 2.5 Pro Experimental (models/gemini-2.5-pro-exp-03-25)
• Updated system prompts to reference Gemini 2.5 Pro
• Improved model usage and documentation

Recent Changes in v0.1.17

• Added list-models command to show all available Gemini models
• Improved error handling for models that don't exist or require permission
• Added model initialization test to verify model availability
• Updated help documentation with new commands

Recent Changes in v0.1.16

• Fixed file creation issues: The CLI now properly handles creating files with content
• Enhanced tool pattern matching: Added support for more formats that Gemini might use
• Improved edit tool handling: Better handling of missing arguments when creating files
• Added special case for natural language edit commands (e.g., "edit filename with content: ...")

Recent Changes in v0.1.15

• Fixed tool execution issues: The CLI now properly processes tool calls and executes Bash commands correctly
• Fixed argument parsing for Bash tool: Commands are now passed as a single argument to avoid parsing issues
• Improved error handling in tools: Better handling of failures and timeouts
• Updated model name throughout the codebase to use gemini-1.5-pro consistently

Known Issues

• If you created a config file with earlier versions, you may need to delete it to get the correct defaults:

  rm -rf ~/.config/gemini-code
  

License

MIT