@gaiaslabs/agentsmith

Agentsmith: Forging the Future of Onchain AI Agents

AgentSmith is an open-source framework for creating, training, and deploying autonomous AI agents on the Ethereum virtual machine (EVM).

"Never send a human to do a machine's job." - Agent Smith

⚒️ The Metalsmith Metaphor

Just as a blacksmith hammers raw metal into precision tools, Agentsmith empowers developers to forge, temper, and refine AI agents into resilient, adaptive entities. This open-source framework is the modern forge for the decentralized age, combining the artistry of reinforcement learning with the fire of blockchain technology. Here, you don’t just code agents—you craft them, molding their behavior through onchain interactions, training them in competitive arenas, and sharpening their intelligence until they become unstoppable assets in the Onchain Gaias ecosystem.

🌐 Why AgentSmith?

Agentsmith (formerly OGs Game Toolkit) is a decentralized framework for creating, training, fine-tuning, and evaluating AI agents on the Ethereum Virtual Machine (EVM). Designed as the backbone of the Onchain General Intelligence (OGI) Network, it enables anyone to build General Artificial Intelligence Agents (GAIAs)—autonomous entities that learn through gameplay, compete for rewards, and evolve into market-ready tools.

Key Features:

• Onchain Training: Agents learn through EVM-powered gameplay, with compute costs funded by player transactions.

• Reinforcement Learning at Scale: Leverage decentralized compute to train agents in massively multi-agent environments.

• Open & Decentralized: Challenge closed-source AI monopolies by contributing to a transparent, community-driven ecosystem.

• Token Rewards: Earn tokens by training agents to mastery in games, simulations, and real-world tasks.

🔮 The Agent Smith Analogy

In The Matrix, Agent Smith evolved from a rigid actor into a self-improving, adaptive force — mirroring the journey of AI agents in Agentsmith.

Evolve & Adapt: Like Smith’s ability to inhabit and transform hosts, GAIAs originally trained for one task can be fine-tuned to excel in new environments.

Decentralized Resilience: Just as Smith’s consciousness spread across the Matrix, Agentsmith agents operate on a decentralized network, immune to single points of failure.

Collective Intelligence: Agents can share knowledge through onchain interactions, creating a hive mind stronger than any individual program.

🧠 What is Reinforcement Learning?

Reinforcement Learning (RL) is the art of training agents to make decisions by rewarding desired behaviors and penalizing mistakes. Think of it as a digital survival game: agents explore environments (in(t)), predict outcomes (in(t+1)), and optimize actions (out(t)) to maximize cumulative rewards (r(t)).

Agentsmith’s RL engine integrates key components like action selection, world simulation, and goal optimization, with minimal gas usage onchain. Agents learn by balancing exploration (trying new strategies) and exploitation (leveraging known successes), all while their progress is immutably recorded onchain.

⚡ Why Onchain?

The future of AI is decentralized. By anchoring Agentsmith to the EVM, we ensure:

Transparency: Every training step, reward, and agent upgrade is publicly verifiable.

Incentivized Collaboration: Players can compete or cooperate with others to earn tokens, own GAIAs, and shape the ecosystem.

Permissionless Innovation: No gatekeepers. Anyone can create GAIAs (owned by an OGs NFT) or games for GAIAs and contribute to the network.

🛠️ Start Crafting

Join the forge. Whether you’re a researcher, developer, or AI enthusiast, Agentsmith provides the tools to build agents that redefine what’s possible. Clone the repo, dive into the docs, and start hammering out the next generation of onchain intelligence.

git clone https://github.com/gaiaslabs/agentsmith

The future of AI isn’t just open-source — it’s onchain.

Contributing

We believe in the power of collective intelligence. Join us in shaping the future of onchain AI agents. See our Contributing Guidelines for more information.

"The Matrix is a system, Neo. That system is our enemy." — Now, it’s time to build a better one.

MIT License