Jemma v2: Your Free Code Assistant

Description: Jemma is a command-line code assistant powered by LLMs. v2 is a complete refactor with a modular architecture — agent loop, pluggable tools, and multi-model support via LiteLLM.

Installation:

pip install .

Configuration:

Before using Jemma, configure your API key:

jemma-init

Alternatively, set the GEMINI_API_KEY environment variable.

Project Structure:

jemma/
├── agent/              # Agent loop orchestration
│   └── loop.py
├── exceptions/         # Custom exceptions
│   ├── file_operation_exceptions.py
│   ├── inference_exceptions.py
│   └── tool_call_exceptions.py
├── services/           # Core services
│   └── tool_service.py
├── tools/              # Pluggable tool definitions
│   ├── file_operations/
│   │   ├── read_file.py
│   │   └── write_to_file.py
│   ├── search_tools/
│   │   └── grep_search.py
│   └── registry.py
├── validators/         # Input validation for tools
│   ├── file_operations_validators.py
│   └── tool_call_validators.py
├── agentic_response.py
├── inference.py
├── message_enum.py
└── response_enums.py
tests/
└── test_tools.py

Tools:

Tool	Description
read_file	Read file content with optional line ranges
write_to_file	Edit existing files by replacing content
write_new_file	Create and write to new files
grep_search	Search for patterns across files and directories

What's New in v2:

• Modular architecture with agent loop, tools, and validators
• Tool registry with schema definitions for LLM function calling
• Multi-model support via LiteLLM
• Streaming inference with recursive tool call execution
• Pluggable tool system — easy to add new tools
• Input validation layer for tool arguments and file operations
• TUI support via Textual

Roadmap

• Complete agent loop with tool call handling
• Add support for more models via LiteLLM
• Improve error handling and logging
• Add unit tests
• Create command for starting new projects
• Add more tools (shell execution, web fetch)