HEAVEN Base Framework

⭐ 0 stars • 🕑 Updated 2026-07-09

Marketplace • Docs

📦 Auto-published from the monorepo • CHANGELOG • sancovp/heaven-framework

Hierarchical, Embodied, Autonomously Validating Evolution Network

HEAVEN Base is the foundational framework for building autonomous AI agents with cross-framework compatibility, event-driven architecture, and self-modifying capabilities. This is the core library that powers the HEAVEN metaprogrammatic agent framework where prompts, tools, agents, and code can all generate each other.

Features

• Cross-Framework Support: Works with both LangChain and Google ADK backends
• Unified Event System: Standardized HEAVEN events for consistent agent communication
• Flexible Agent Architecture: Build custom agents by extending BaseHeavenAgent
• Tool System: Create reusable tools by extending BaseHeavenTool
• History Management: Advanced history tracking with event extraction
• Multiple LLM Providers: Support for OpenAI, Anthropic, Google, DeepSeek, Groq, and more

Installation

pip install heaven-framework

Toolbox MCP

Claude Code

"heaven-framework-toolbox": {
      "type": "stdio",
      "command": "python",
      "args": [
        "-m",
        "heaven_base.mcps.toolbox_server"
      ],
      "env": {
        "HEAVEN_DATA_DIR": "YOUR_HEAVEN_DATA_DIR", # /tmp/heaven_data recommended
        "OPENAI_API_KEY": "sk-YOUR_OPENAI_API_KEY"
      }
    }

SDK

Quick Start

Creating a Simple Agent

from heaven_base import BaseHeavenAgent, HeavenAgentConfig

class MyAgent(BaseHeavenAgent):
    @classmethod
    def get_default_config(cls) -> HeavenAgentConfig:
        return HeavenAgentConfig(
            name="MyAgent",
            system_prompt="You are a helpful assistant.",
            model="gpt-4",
            temperature=0.7
        )

# Run the agent
agent = MyAgent()
result = await agent.run("What is the meaning of life?")
print(result)

Creating a Custom Tool

from heaven_base import BaseHeavenTool, ToolArgsSchema

class CalculatorToolArgsSchema(ToolArgsSchema):
    arguments = {
        'expression': {
            'type': 'str',
            'description': 'Mathematical expression to evaluate',
            'required': True
        }
    }

class CalculatorTool(BaseHeavenTool):
    name = "calculator"
    description = "Evaluates mathematical expressions"
    args_schema = CalculatorToolArgsSchema
    
    def _run(self, expression: str) -> str:
        try:
            result = eval(expression)
            return f"The result is: {result}"
        except Exception as e:
            return f"Error: {str(e)}"

Core Components

• BaseHeavenAgent: Base class for all HEAVEN agents
• BaseHeavenTool: Base class for all HEAVEN tools
• UnifiedChat: Multi-provider LLM interface
• HeavenEvent: Standardized event format
• History: Advanced conversation history management

Development Roadmap

HEAVEN Base is actively evolving with a clear roadmap to become a complete agent development toolkit:

🚀 Core Infrastructure (Current)

• ✅ BaseHeavenAgent & BaseHeavenTool - Foundation classes for agents and tools
• ✅ Cross-Framework Support - LangChain and Google ADK compatibility
• ✅ HEAVEN Events - Standardized event system for agent communication
• ✅ Registry System - Data storage and retrieval with cross-registry references
• ✅ History Management - Conversation tracking with ~/.heaven/ user directory

🛠️ Local Execution (In Progress)

• Hermes Local Execution - Non-containerized agent orchestration (replaces complex Docker setup)
• Core Default Tools - BashTool, NetworkEditTool, SafeCodeReaderTool out of the box
• Simple Agent Runner - run_agent() utility for immediate agent execution

🧠 LLM Integration Improvements

• LiteLLM Integration - Replace LangChain internals with LiteLLM for better provider support
• Unified Message Formats - Standardize all message handling to OpenAI format
• Enhanced Provider Support - Better error handling and feature parity across providers

🎯 Prompt Engineering System

• Prompt Injection System - Build input prompts from reusable blocks or freestyle strings
• Template Management - Organize and version prompt templates
• Dynamic Composition - Programmatically construct complex prompts

🔄 Context Engineering

• Weave Operations - Extract and reorganize conversation history
• Inject Operations - Add files, context, or arbitrary content to conversations
• Context Compression - Manage long conversations with smart truncation

🏗️ Workflow Orchestration

• HeavenWorkflow Class - Turn all components into LangGraph nodes
• Lego-Style Building - Compose agents, tools, and operations as reusable blocks
• Visual Workflow Design - Drag-and-drop agent workflow construction
• Hierarchical Execution - Nested workflows with state management

🎯 The Vision

HEAVEN Base aims to be the complete toolkit for agent development:

• Install → pip install heaven-framework
• Run → hermes.run_agent(agent, prompt)
• Compose → Build prompts from blocks and inject context
• Orchestrate → Create complex workflows as LangGraph nodes
• Scale → Hierarchical agent networks with automatic state management

Documentation

For full documentation, visit https://heaven-base.readthedocs.io

License

MIT License - see LICENSE file for details.