As AI agents take on more responsibility in software development and financial transactions, the digital asset infrastructure company is introducing tools to keep their actions within predefined limits.
AI agents are increasingly capable of writing software, executing commands and handling financial transactions. But as businesses give these systems more autonomy, a new challenge is emerging: how to make sure they don't take actions their owners never intended.
AEREDIUM, a digital asset infrastructure company, is addressing that problem with two products designed to put controls between AI agents and the systems they operate.
The products, called The Chair and MCP Connector, are available through the company's new aeredium.ai platform. Both use a policy engine that evaluates an AI agent's proposed actions before allowing them to proceed.
Rather than giving an agent unrestricted access to software systems or digital assets, organizations can establish rules determining which actions are permitted, which are blocked and which require human approval.
Each decision is digitally signed and recorded, creating a verifiable history of what the agent attempted and how the system responded.
“AI agents are no longer just answering questions. They're writing code, moving money and making decisions that have real consequences,” says Albert Dadon, CEO and Founder of AEREDIUM. “The challenge isn't just making these agents more capable, but making sure they operate within the rules we set. With The Chair and MCP Connector, we're giving organizations a way to put AI to work without giving up control over what it can do.”
The first product, The Chair, focuses on software development, where AI agents can already perform tasks that traditionally required direct human involvement.
The tool sits between an AI coding agent and an organization's development environment, checking actions such as modifying files, executing commands, merging code and pushing changes against company policies.
Actions that meet established rules can proceed automatically, while those that violate the rules are denied. Requests requiring additional authorization are sent to a control panel for human review.
The system also produces signed records showing what an agent attempted, which policy was applied and who authorized the action.
The Chair supports Anthropic's Claude Code through a dedicated plugin and provides a governed shell for other coding agents, including those developed by OpenAI. Companies can configure their policies through a questionnaire instead of writing the rules directly in code.
AEREDIUM is offering the product with a 30-day free trial.
The second product, MCP Connector, applies similar controls to financial activities.
It combines a digital asset wallet with a policy engine, allowing AI agents to execute payments or trades without giving them unrestricted control over funds.
Each agent receives its own wallet, a defined set of permissions and a designated human owner. That owner approves the agent's mandate using a passkey and can halt or delete the agent at any time.
AEREDIUM currently supports two types of financial agents.
A Payer agent can transfer stablecoins, including USDC, USDT, PYUSD and EURC, to addresses approved by its owner, but cannot execute trades.
A Trader agent can trade on designated exchanges using its own wallet, but cannot transfer funds to external recipients.
The distinction is intended to separate payment authority from trading authority, reducing the range of financial actions any individual agent can perform.
According to AEREDIUM, the wallets use threshold signatures generated inside secure enclaves, so a complete private key is never created and the company does not hold a key to the owner's funds.
MCP Connector currently supports Ethereum, Arbitrum and Base, with transaction records retained for seven years. It is designed for Claude and is open to integration with OpenAI and other AI frameworks.
The two products represent an expansion of AEREDIUM's existing digital asset infrastructure into AI agent governance.
The company has previously focused on institutional custody and blockchain settlement, using cryptographic technology and predefined policies to control how transactions are authorized.
Its new AI products extend that approach beyond financial infrastructure to autonomous software operations and agent-driven transactions.
The underlying idea is that an AI agent's ability to perform a task should not automatically give it permission to do so. Instead, actions must satisfy rules established by the organization or individual responsible for the agent.
That distinction could become increasingly relevant as AI systems move beyond generating responses and begin interacting directly with business software, financial accounts and development environments.
For AEREDIUM, the launch marks an effort to apply institutional-style controls to a new category of automated activity, one in which software agents are increasingly being trusted to act on behalf of humans.
The Chair and MCP Connector launched on Oct. 13, and are available through AEREDIUM's AI platform.