brocode 0.1.3

brocode is a lightweight coding assistant in CLI

BroCode

Goal

BroCode is a study of agentic workflow and AI agents framework designed to solve coding problems. It's model-agnostic and aims to work with all files in a repository like Claude, Amazon Q Developer, and other AI coding tools. Future versions will include code acceptance/rejection capabilities.

Overview

A CLI tool for managing and running LLM-based chat agents with support for multiple model backends. BroCode uses an agentic workflow system that combines code generation, chat capabilities, and codebase analysis in an interactive flow.

Installation

Use this:

pip install brocode  

Or this:

uv add brocode  

Quick Start

1. Register a model:

brocode register --path mylocal.py --model llama3.2-11b --default

2. Start chatting:

brocode start

Commands

Register Models

Register LLM models from Python files:

# Register and set as default
brocode register --path mylocal.py --model mymodel --default

# Register with auto-generated name
brocode register --path mylocal.py

Start Chat

Start interactive chat sessions:

# Use default model
brocode start

# Use specific model
brocode start --llm mymodel

Model Management

# List registered models
brocode model list

# Remove models interactively
brocode model remove

Creating Custom Models

Create a Python file with your LLM class:

# mylocal.py
from brollm import BaseLLM, BedrockChat
from brocode.register import register_llm

@register_llm("llama3.2-11b")
class MyLocalLLM(BedrockChat):
    def __init__(self):
        super().__init__(model_name="us.meta.llama3-2-11b-instruct-v1:0")

How BroCode Works

When you run brocode start, BroCode creates a brosession directory in your current location and initiates an agentic workflow with two main modes:

BroSession Directory Structure

BroCode organizes all session-related files in a brosession directory:

your-project/
├── brosession/
│   ├── brocode_config.yaml    # Model configurations
│   ├── session.db             # Session data
│   └── prompt_hub/            # Customizable prompts
│       ├── chat.md           # Chat assistant persona
│       └── code_generator.md  # Code generation guidelines
└── your-code-files...

Session Management

• Per-Directory Sessions: Each directory gets its own brosession with independent configurations
• Customizable Prompts: Edit files in brosession/prompt_hub/ to customize AI behavior
• Portable Sessions: Move or copy brosession folders to share configurations
• Easy Cleanup: Delete brosession folder to reset everything

Workflow Overview

[Start] → [Setup BroSession] → [User Input] → [Route Decision]
                                     ↓
                             ┌─────────────────┐
                             ↓                 ↓
                        [Code Mode]      [Chat Mode]
                             ↓                 ↓
                        [Code Generator]  [Chat Agent]
                             ↓                 ↓
                             └─────────────────┘
                                     ↓
                             [Back to User Input]

Interactive Commands

• /agents - Enter agent selection mode with interactive menu

  • Coder Agent: Full-featured code operations with create and modify capabilities
    • Create: Generate new Python code with codebase reference and multi-file selection
    • Modify: Update existing Python code with preview-before-apply workflow
  • Analyst Agent: Data analysis capabilities (Coming in next release)

• /exit - Quit the session

• /clear - Clear chat history

• Default input - Enter chat mode for general conversation

Coder Agent Features

Interactive CRUD Operations

• Arrow Key Navigation: Use ↑↓ keys to navigate all menus
• Multi-Select File Reference: When referencing codebase:
  • Select multiple files with Space bar
  • "✅ Select All" option for bulk selection
  • Visual file tree with 📄 icons
• Visual Code Display: All code shown with syntax highlighting and line numbers
• Cross-Platform Clipboard: Copy generated code to clipboard (when pyperclip available)

Create Workflow

1. Task Description: What you want to create
2. Codebase Reference (optional):
   • y/n prompt for existing codebase
   • Folder path input
   • Multi-select files for reference
3. Output Method: Display in terminal or save to file
4. Code Generation: Always displays code first, then saves if requested
5. Clipboard Copy: Optional copy to clipboard

Modify Workflow

1. Target File Selection: Choose Python file to modify
2. Modification Request: Describe changes needed
3. Codebase Reference (optional): Reference additional files for context
4. Preview Changes: View proposed modifications before applying
5. Accept/Reject: Choose to apply changes or cancel
6. No Backup Files: Direct modification without .backup files

Code Generation & Modification Features

• Rich Visual Display: All code shown with syntax highlighting, line numbers, and colored panels
• AST-Based Analysis: Intelligent parsing of existing Python codebases
• Multi-File Context: Reference multiple files simultaneously for better context
• Interactive File Selection: Visual file browser with multi-select capabilities
• Clipboard Integration: Copy generated code directly to system clipboard
• Preview-Before-Apply: See modifications before they're applied to files
• Error Handling: Comprehensive error handling with user choices (retry/skip/cancel)
• File Support: Works with both .py and .md files for codebase reference
• PEP 8 Compliance: Automatic style guideline following
• Google Docstrings: Standardized documentation format
• Type Hints: Full type annotation support

Chat Mode

Provides general coding assistance, debugging help, and technical discussions using a knowledgeable coding assistant persona.

Configuration

Models are stored in brosession/brocode_config.yaml in your current directory. Use brocode model config to see the exact location.

Configuration File Structure

models:
  llama3.2-11b: /path/to/mylocal.py
  gpt-4: /path/to/gpt4_model.py
default_model: llama3.2-11b

Customizing Prompts

After running brocode start once, you can customize the AI behavior by editing:

• brosession/prompt_hub/chat.md - Chat assistant personality and instructions
• brosession/prompt_hub/code_generator.md - Code generation guidelines and style
• brosession/prompt_hub/code_modifier.md - Code modification guidelines and behavior

Smart File Management: BroCode only copies missing prompt files, preserving your customizations while ensuring all required files are present.

Dependencies

• Python >=3.12
• click >=8.2.1
• brollm >=0.1.2
• broflow >=0.1.4
• broprompt >=0.1.5
• inquirerpy >=0.3.4
• rich >=14.1.0
• pyfiglet >=1.0.4