New Open Source Tool From CoinFello and MetaMask Brings Secure Onchain Transactions to AI Agents
CoinFello and MetaMask launch an open-source tool enabling AI agents to execute secure onchain transactions using delegated permissions.
The convergence of artificial intelligence and blockchain infrastructure continues to accelerate as developers search for ways to allow autonomous systems to interact safely with financial networks. CoinFello, an AI agent platform focused on secure smart contract interaction, has introduced a new open-source capability designed to make that vision more practical while preserving user control over digital assets.
The company unveiled the OpenClaw skill in collaboration with MetaMask, a development aimed at enabling AI agents to perform on-chain transactions using delegated permissions rather than direct access to private keys. The framework allows agents such as CoinFello’s Moltbots to execute blockchain operations while the signing keys remain stored on the user’s device.
Unlike some experimental AI wallet models that rely on centralized trusted execution environments or directly hold private keys, CoinFello’s system attempts to reduce risk through a permission-based architecture. The framework uses ERC-4337 smart accounts combined with ERC-7710 delegation mechanisms available through the MetaMask Smart Accounts Kit, allowing users to grant narrowly defined permissions to automated agents.
In practice, the platform translates natural language instructions into blockchain transactions. When a user issues a request, CoinFello’s agent interprets the command, generates the corresponding transaction and submits it for verification before execution. According to the company, this process follows a least-privilege model that minimizes the amount of authority granted to autonomous systems.
This design reflects growing industry concern about the security implications of giving AI agents unrestricted access to digital wallets. As the idea of “agentic” software capable of managing financial tasks gains traction, developers are increasingly exploring methods that allow automation without sacrificing custody or control.
CoinFello’s leadership argues that stronger safeguards will be essential if AI agents are to become meaningful participants in blockchain ecosystems. Brett Cleary, the company’s chief technology officer, said the sector must move beyond architectures that expose sensitive keys to autonomous software.
“The CoinFello Skill introduces hardware-isolated keys and fine-grained delegations, giving AI agents a secure way to execute transactions while helping bootstrap onchain capabilities for the broader agent ecosystem,” Cleary said.
The launch arrives amid rapid expansion within the OpenClaw development community, an open framework designed for building AI agent tools that interact with decentralized systems. The project has seen substantial developer interest, accumulating more than 150,000 stars on GitHub along with tens of thousands of forks and hundreds of thousands of recent downloads through npm.
With the new skill, Moltbots can carry out a wide range of blockchain operations triggered through conversational prompts. These include ERC-20 token swaps, cross-chain bridging, NFT interactions, staking, lending and automated portfolio management. More advanced functions, such as multi-step trading strategies, can also be executed through sequences of instructions interpreted by the AI system.
The tool follows the Agent Skills specification used across OpenClaw environments and is compatible with development tools such as Claude Code. Released under the MIT open-source license, it allows developers to adapt, modify and integrate the capability into their own AI-driven blockchain applications.
CoinFello says the system is designed to support broader interoperability within the emerging ecosystem of autonomous on-chain agents. Moltbots can delegate permissions to any compatible blockchain agent, allowing multiple automated systems to collaborate while still respecting user-defined limits.
Future development will focus on expanding the permission framework and deepening integration with the MetaMask Smart Accounts Kit. As the boundaries between artificial intelligence and decentralized finance continue to blur, tools like these may help determine whether automated agents can safely operate in the financial networks of tomorrow.
