broadie 1.0.5

A powerful multi-agent framework with persistence, API server, and agent-to-agent communication

Broadie

A powerful multi-agent framework with persistence, API server, and agent-to-agent communication.

Features

• Agent & SubAgent System: Create intelligent agents with specialized sub-agents
• Multiple Model Support: Google Gemini and Vertex AI integration
• Persistence: SQLite and PostgreSQL backends with vector storage
• API Server: FastAPI-based REST and WebSocket endpoints
• Agent-to-Agent Communication: Secure peer discovery and function invocation
• Tool System: Extensible tool registry for agent capabilities
• Memory Management: Persistent memory with semantic search
• CLI Tools: Command-line interface for agent management
• Notifications: Slack integration for alerts and updates

Getting Started

Follow these steps to set up and run Broadie locally.

Prerequisites

• Python 3.10 or newer
• pip
• git

Setup

git clone https://github.com/your-org/broadie.git
cd broadie
python3 -m venv venv
source venv/bin/activate        # On Windows: venv\Scripts\activate
pip install -e .[dev]

Alternatively, install the latest release from PyPI:

pip install broadie

Configuration

Copy the example environment file and update with your settings:

cp .env.example .env

Quick Start

Installation

pip install broadie

Basic Usage

1. Configuration-driven Agent

Create main.json (minimal, tool-less agent configuration):

{  
  "name": "my_agent",
  "description": "A helpful assistant",
  "instruction": "You are a helpful AI assistant.",
  "model": {"provider": "google", "name": "gemini-2.0-flash"}
}

Run the agent:

broadie run main.json

2. Programmatic Agent

from broadie import Agent, SubAgent, tool

# Define a custom tool
@tool(name="analyze", description="Analyze data")
def analyze(data: str) -> str:
    return f"Analysis of: {data}"

# Sub-Agent defines its own configuration
class AnalysisAgent(SubAgent):
    def build_config(self):
        return {
            "name": "analysis_agent",
            "description": "Performs data analysis tasks",
            "instruction": "You are a data analysis specialist."
        }

# Main Agent coordinates sub-agents
class CoordinatorAgent(Agent):
    def build_config(self):
        return {
            "name": "coordinator",
            "description": "Coordinates tasks",
            "instruction": "You break tasks among sub-agents.",
            "sub_agents": [AnalysisAgent()]
        }

# Use the agent
agent = CoordinatorAgent()
response = await agent.process_message("Analyze this data: sample")
print(response)

Configuration

Environment Variables

Copy .env.example to .env and configure:

# Google AI
GOOGLE_API_KEY=your_api_key_here
DEFAULT_GEMINI_MODEL=gemini-2.0-flash

# Agent-to-Agent
A2A_ENABLED=true
A2A_AGENT_ID=agent-12345
A2A_AGENT_NAME=my_agent
A2A_TRUSTED_AGENTS=agent-456,agent-789

# Database
DB_DRIVER=sqlite
DB_NAME=broadie.db

# Slack (optional)
SLACK_BOT_TOKEN=xoxb-your-token
SLACK_CHANNEL=#alerts

Agent Configuration

Agent configurations are JSON files defining agent identity and model settings. These configs are minimal and do not include custom tools; to extend with tools, use the programmatic API.

{
  "name": "agent_name",
  "description": "Agent description",
  "instruction": "System prompt for the agent",
  "model": {
    "provider": "google",
    "name": "gemini-2.0-flash"
  },
  "model_settings": {
    "temperature": 0.1,
    "max_tokens": 8000
  },
  "sub_agents": ["subagent1.json", "subagent2.json"]
}

CLI Commands

# Initialize new project (scaffolds .json configs, agent.py, .env.example, README.md)
broadie init my_project

# Run agent from config or code
broadie run main.json

# Start API server with A2A heartbeat (reads settings from .env)
broadie serve main.json
# alternative: serve a programmatic agent
broadie serve module.path:AgentClass

# Create a minimal JSON agent config in agents/ directory
broadie create agent <agent_name>

# Create a minimal JSON sub-agent config in agents/ directory
broadie create sub-agent <subagent_name>

API Server

Start the API server by pointing to a JSON config or a Python agent reference:

# Serve JSON-configured agent
broadie serve main.json

# Serve programmatic agent from code
broadie serve module.path:AgentClass

Endpoints

• GET /health - Health check
• POST /agents/{id}/invoke - Invoke agent function
• GET /threads - List all conversation threads
• GET /threads/{thread_id}/messages - Get messages for a specific thread
• WebSocket /ws - Real-time chat interface

Example API Usage

import httpx

# Invoke agent
response = httpx.post(
    "http://localhost:8000/agents/my_agent/invoke",
    json={"message": "Hello, agent!"}
)
print(response.json())

Agent-to-Agent Communication

Agents can discover and communicate with each other:

# Agent A can invoke functions on Agent B
result = await agent_a.invoke_peer_agent(
    "agent_b_id", 
    "analyze_threat",
    {"threat_data": "suspicious_activity"}
)

Security Model

• Agents can see all peers via registry
• Only trusted agents (configured in A2A_TRUSTED_AGENTS) can invoke functions
• All communications include agent identity verification

Memory System

Agents have persistent memory with semantic search:

# Store memory
memory_id = await agent.memory_manager.remember(
    "Important information",
    importance=0.8,
    tags=["important", "project_x"]
)

# Search memories
memories = await agent.memory_manager.search(
    "project information",
    limit=5
)

# Recall specific memory
memory = await agent.memory_manager.recall(memory_id)

Tools

Create custom tools for agents:

from broadie.core.tools import tool

@tool(name="weather", description="Get weather information")
async def get_weather(location: str) -> str:
    # Weather API integration
    return f"Weather in {location}: Sunny, 72°F"

# Register tool
from broadie.core.tools import get_global_registry
registry = get_global_registry()
# Tool is auto-registered via decorator

Development

Setup Development Environment

git clone https://github.com/your-org/broadie
cd broadie
pip install -e ".[dev]"

Running Tests

pytest
pytest --cov=src/broadie --cov-report=html

Code Quality

black src/ tests/
isort src/ tests/
flake8 src/ tests/
mypy src/

Architecture

Broadie follows SOLID principles with clean separation of concerns:

• Core: Agent, SubAgent, Tool, Memory abstractions
• Config: Environment and JSON configuration management
• Persistence: SQLite, PostgreSQL, and vector storage backends
• API: FastAPI server with REST and WebSocket support
• A2A: Agent-to-agent communication and discovery
• CLI: Command-line tools for agent management
• Utils: Logging, exceptions, and schema validation

Contributing

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing-feature)
3. Make your changes
4. Add tests for your changes
5. Run the test suite (pytest)
6. Commit your changes (git commit -m 'Add amazing feature')
7. Push to your branch (git push origin feature/amazing-feature)
8. Open a Pull Request

License

This project is licensed under the MIT License - see the LICENSE file for details.

Support

• Documentation: https://docs.broadie.ai
• Issues: https://github.com/your-org/broadie/issues
• Discussions: https://github.com/your-org/broadie/discussions

Developer Quick Reference

The following notes are adapted from the internal CLAUDE.md guide for convenience.

Testing Commands

• Lint: python -m black --check . or python -m ruff check .
• Type checking: python -m mypy broadie/
• Tests: python -m pytest
• Build: python -m build or pip install -e .

Available Tools

Built-in Tools (Always Available)

• write_todos - Create and manage task lists
• ls - List files in filesystem
• read_file - Read file contents
• write_file - Write content to files
• edit_file - Make exact string replacements

Google Drive Tools (Auto-included when configured)

• search_google_drive - Search files and folders by name/content
• read_google_drive_file - Read document/sheet content
• write_google_drive_file - Create/update documents
• create_google_sheet - Create new spreadsheets
• update_google_sheet - Update sheet data
• list_drive_folders - List folders and contents

Slack Tools (Auto-included when configured)

• search_slack_messages - Search messages across workspace
• send_slack_message - Send messages to channels
• send_slack_dm - Send direct messages
• list_slack_channels - List available channels
• list_slack_users - List workspace users
• get_slack_user_info - Get user details
• create_slack_thread - Reply to messages
• upload_slack_file - Upload files to channels

Architecture Notes

• All tools are automatically available — no need to specify them in agent configs
• Built-in tools are always included (write_todos, file operations, etc.)
• Integration tools are automatically included when properly configured
• Tools use a global registry with auto-registration
• Configuration managed through broadie.config.integrations
• Tool registry supports categories: builtin, google_drive, slack
• Integration tools gracefully fail when dependencies are unavailable
• Production-quality error handling with helpful setup instructions

Agent Configuration

Agents automatically get all available tools. You can still specify explicit tools in the tools list to add additional custom tools, but built-ins and integrations are always included.

Packaging & Release

Follow these steps to produce installable distributions and publish a release.

1. Build the UI (optional but recommended for releases)

   • cd broad-ui
   • npm install
   • npm run build
   • This outputs the SPA to broadie/static/ui/, which is included in the package.

2. Build Python distributions

   • Ensure build tools are available: pip install -U build twine
   • Or install dev extras: pip install -e .[dev]
   • Build: python -m build
     • Produces dist/broadie-.tar.gz and dist/broadie--py3-none-any.whl

3. Verify contents

   • Inspect the wheel/sdist (tar -tf dist/*.tar.gz) for:
     • broadie/static/ui/**/* (UI assets)
     • broadie/templates/*.json
     • broadie/py.typed

4. Upload to PyPI (or TestPyPI)

   • TestPyPI: twine upload -r testpypi dist/*
   • PyPI: twine upload dist/*

Notes

• Optional extras:
  • dev: tooling for tests and builds (pytest, black, isort, mypy, ruff, build, twine)
  • google, slack, postgres, sqlite: install integration-specific dependencies
• Source distribution is governed by MANIFEST.in; wheels include package-data via pyproject.toml.

Troubleshooting installed CLI

• Symptom: ModuleNotFoundError: No module named 'broadie' right after running broadie.
  • Cause: You may have an older wheel that didn’t include the package modules.
  • Fix: Upgrade to the latest version: python -m pip install -U broadie.
  • If testing from a local build, ensure the wheel includes the broadie/ package:
    • Build: python -m build
    • Inspect wheel: python -m pip install wheel-inspect && wheel2zip dist/*.whl && unzip -l dist/*.whl.zip | sed -n '1,120p'
    • You should see entries under broadie/.... Rebuild if missing.