Open Source Agent Frameworks
Agent frameworks provide building blocks for creating software agents that use language models to interpret requests, call tools, maintain state, and perform multi-step tasks. They help developers manage the coordination, memory, and execution logic that would otherwise need to be assembled from scratch, while making agent behavior easier to test and adapt. Some frameworks focus on individual agents, while others support workflows involving multiple agents or external services.
Open source options range from lightweight libraries and orchestration runtimes to visual builders, self-hosted platforms, and tools for testing or governing agent behavior. When choosing one, consider project maturity, license, maintenance activity, supported model and tool integrations, deployment requirements, and the complexity of its abstractions. These frameworks are useful to developers and teams prototyping agent applications, building production workflows, or seeking greater control over how AI systems run and connect to other software.
13 repositories · updated October 2, 2026
Shepherd: Reversible Execution Traces for Programmable Meta-Agents
Shepherd is a Python runtime substrate designed for agent work requiring inspection, reversibility, and supervision. It records agent runs as durable, inspectable execution traces, enabling meta-agents to observe, fork, replay, and revert any operation. This framework couples agents and environments using a copy-on-write fork, offering significant performance benefits and robust permission enforcement.
HarnessRouter: Unified Interface for AI Agent Harnesses
HarnessRouter Community Edition provides a self-hosted, Apache-2.0 licensed unified interface for various AI agent harnesses like Codex, Claude Code, and Hermes. It allows users to run multiple agents through a single API, offering features such as sessions, streaming, file handling, and cancellation. The project implements the open-standard Unified Harness Protocol (UHP), ensuring users maintain control over their keys and infrastructure.
AgentsKit: The Complete JavaScript Toolkit for Building AI Agents
AgentsKit is a comprehensive JavaScript toolkit designed for building AI agents, offering a lightweight core and a modular ecosystem. It provides essential components like UIs, autonomous runtime, tools, memory, and RAG, enabling developers to create sophisticated agents from simple chat interfaces to complex autonomous systems. This framework aims to simplify agent development by offering composable parts and avoiding the need to glue multiple incompatible libraries together.
Declarative Agents: Profile-Driven LLM Agent Runtime in Go
Declarative Agents by Nokia Bell Labs offers a profile-driven runtime and design patterns for building tool-augmented LLM agents. It allows defining agents, their tools, states, and transitions via YAML profiles, eliminating the need for code changes for workflow alterations. This Go-based framework promotes flexible and dependable agent development.
Ralph Orchestrator, An Advanced Framework for Autonomous AI Agent Orchestration
Ralph Orchestrator is a robust, Rust-based framework designed for autonomous AI agent orchestration. It implements the innovative "Ralph Wiggum technique," a methodology focused on continuous iteration to ensure AI agents complete complex tasks effectively. This powerful tool supports multiple AI backends and offers features like a "hat system" for specialized personas and human-in-the-loop interaction via Telegram.
AFT: A Sensorimotor Cortex for Coding Agents with IDE and OS Capabilities
AFT provides coding agents with advanced IDE and OS capabilities, acting as a sensorimotor cortex for perception and action within a codebase. It enhances agent productivity and reduces token usage by offering structured code perception, precise symbol-aware edits, and background task management. Part of the CortexKit family, AFT integrates with platforms like OpenCode and Pi to elevate agent interaction with code.
Agent Governance Toolkit: Policy Enforcement for Autonomous AI Agents
The Microsoft Agent Governance Toolkit (AGT) provides robust policy enforcement, zero-trust identity, and execution sandboxing for autonomous AI agents. It addresses critical security and compliance challenges, ensuring agents operate within defined boundaries and providing tamper-evident audit trails. AGT covers all 10 items of the OWASP Agentic Top 10, making it essential for shipping AI agents to production securely.
taOS: Self-Hosted AI Agent OS for Your Hardware and Data Sovereignty
taOS is a self-hosted AI agent operating system designed to keep your AI's memory, chats, agents, and files on your own hardware, prioritizing privacy and data sovereignty. It offers a full web desktop, a multi-framework group chat, and the ability to auto-cluster various consumer hardware like Raspberry Pi, Mac mini, and gaming PCs into a distributed AI compute mesh. This platform empowers users with local-first AI capabilities, offline by default and cloud by choice.
awesome-ai-agents: A Curated List of AI Agent Resources
awesome-ai-agents is a comprehensive, curated list of resources for building and understanding AI agents. It covers frameworks, tools, platforms, research papers, and more, making it an essential guide for anyone exploring the rapidly evolving field of autonomous AI systems.
nanobot: An Ultra-Lightweight, Self-Hosted Personal AI Agent Framework
nanobot is an ultra-lightweight, open-source, self-hosted personal AI agent framework built in Python. It offers a WebUI, tools, memory, multi-agent workflows, and automation capabilities, making it a versatile solution for personal AI tasks. Users can deploy it across various platforms, including chat apps, for seamless integration.
Deep Agents: The Batteries-Included Agent Harness for Complex AI Tasks
Deep Agents is an agent harness built on LangChain and LangGraph, designed to simplify the creation of complex AI agents. It comes equipped with essential tools like planning, filesystem access, and the ability to spawn sub-agents, enabling it to handle sophisticated agentic tasks out of the box. This framework provides a ready-to-run agent that can be easily customized with additional tools, models, and prompts.
mcp-agent: Build Effective AI Agents with Model Context Protocol in Python
mcp-agent is a powerful Python framework designed to help developers build effective AI agents using the Model Context Protocol (MCP) and simple, composable workflow patterns. It fully implements MCP, providing robust support for agent lifecycle management and integrating patterns from Anthropic's 'Building Effective Agents'. This framework simplifies the creation of durable, production-ready agent applications.