Ledger SAS today launched Ledger Agent Stack, an open-source toolkit that lets developers build artificial intelligence agents that can read balances and draft crypto transactions but cannot move funds without a human signing off on hardware.

The Paris-based company, maker of the widely used Ledger hardware wallets, is pitching the release as an answer to a basic problem in agentic AI. Agents can suggest a swap or prepare a payment, but letting them act on their own has meant handing software the keys. Ledger Agent Stack keeps those keys on a physical device. Its design rule is that agents propose, humans approve and the Ledger signer enforces.

Sensitive actions have to be confirmed on a Ledger signer before they execute. Read-only requests such as checking a balance or reviewing history run on their own. The company said the split means that even if an agent’s software environment is compromised, funds cannot move and secrets cannot be reached without a physical confirmation.

The kit ships as a set of open-source components. Device Management Kit Skills are instruction sets that teach any large language model runtime how to wire Ledger hardware into a workflow without rebuilding wallet logic. The Ledger Wallet CLI gives agents programmatic access to check balances, receive funds and prepare sends, swaps and staking. The Ledger Enterprise CLI and Ledger Enterprise Multisig CLI extend the same controls to institutional treasury teams, with policy enforcement anchored in hardware security modules.

Two new device apps push the hardware past crypto. OpenPGP encrypts agent secrets so they cannot be read without a Ledger signer plugged in. Security Key gates log in to services including GitHub, npm, 1Password and Discord behind the device, so a stolen password alone is not enough to get in.

“Crypto wallets have protected billions on this standard for years,” said Ian Rogers, chief human agency officer at Ledger. “Ledger Agent Stack allows your agent to use these wallets just as easily as humans.”

The launch leans on a security argument. Ledger cited research putting human error at the root of roughly 60% of breaches and a separate study finding that about 26% of agent skills carry at least one vulnerability. Anchoring every signing decision in hardware, the company said, contains the damage when the software around an agent fails.

Partners including MoonPay Inc. and Shisa Inc. have already used the Device Management Kit to add Ledger support to their products.

Ledger Agent Stack is the first piece of a 2026 AI roadmap the company laid out earlier this year. Later releases, covering agent identity, policies and a “proof of human” check, are due through the rest of the year. Documentation is available at developers.ledger.com/agent-kit.

Founded in 2014, Ledger says it has sold more than 8 million of its signers across more than 165 countries.

Image: Ledger