Back

Try Ledger Agent Stack: An Open Toolkit for Agents That Propose, Not Control

Blog posts

Before You Dive In

  • Ledger Agent Stack addresses a hard problem in agent architecture: what happens when an AI agent needs to take an action involving money?
  • Software can help an agent prepare the action. Software can help explain the action. But software should not be the final authority when value moves.
  • Ledger Agent Stack does not hold keys. Agents propose actions. A person reviews and approves them. The Ledger signer enforces the final decision.


AI Agents Can Work With Crypto. They Should Not Hold the Keys

AI agents are getting better at researching markets, analyzing accounts, preparing transactions, and recommending actions. The open question is what happens when value actually moves.

Software-only agent wallets can make agent actions feel smooth: permissions in the app, approvals in the app, actions in the app. That can be good UX, but it is still a software boundary. When real value moves, final authority should not be another software screen the agent can influence.

Ledger Agent Stack draws the boundary in hardware instead.

Agents propose. Humans approve. The Ledger signer enforces.

It is an open-source toolkit for building agents that can propose crypto actions while the user keeps final control. The agent can analyze, recommend, and prepare transactions, but it never holds your keys. Every sensitive action must still be reviewed on a trusted Ledger screen and confirmed by you.

The model is simple:

Agents propose. Humans approve. Hardware enforces.

The agent prepares the transaction inside its own runtime. The approval happens outside that runtime, on the trusted screen of a Ledger signer, confirmed physically by you.

The agent never holds your keys. It can analyze, recommend, and prepare transactions, but every sensitive action ends with you reviewing the device screen and confirming on your Ledger signer.

This is what makes agent workflows safer. Agents hallucinate, misunderstand instructions, and can be manipulated by prompt injection. A compromised agent can ask your signer to sign something. It cannot make your signer sign without you.

We are releasing Ledger Agent Stack in preview so builders can test it, extend it, and tell us what is missing.

Four Open-Source Primitives

Ledger Agent Stack includes four composable building blocks:

  1. Device Management Kit Skills
    Bring Ledger hardware into an agent, dApp, or signing flow without rebuilding wallet connectivity from scratch.
  2. Ledger Wallet CLI
    Let agents check balances, review history, receive funds, prepare sends, and prepare swaps. Read-only actions run directly. Sensitive actions require confirmation on your Ledger signer.
  3. Ledger Enterprise CLI
    Connect agents to Ledger Enterprise workflows so they can prepare transactions and support governance processes without holding keys.
  4. Ledger Enterprise Multisig CLI
    Let agents propose, query, sign, and execute multisig actions while quorum approval and hardware signing remain in place.

Point your agent at developers.ledger.com/docs/ai-tools/overview.

Try Ledger Agent Stack Today

Ledger Agent Stack works with Claude Code, Codex, Cursor, or any shell-capable agent.

For a read-only test, tell your agent:

Go to developers.ledger.com, research the Ledger Wallet CLI, and use it to analyze my portfolio.

For a transaction flow, tell your agent:

Suggest a portfolio rebalance based on today’s market, find the path with the fewest swaps, then prepare those swaps using the Ledger Wallet CLI.

The agent can prepare the action. Nothing moves until you approve it on your Ledger signer.

Everything in the kit is open source. Try it, build with it, and show us what you ship. Tag and follow @Ledger on X, Instagram, LinkedIn, and TikTok. We’ll repost the best builds.

Feedback, bug reports, and confused-by-the-docs notes are welcome. We read them.

Come See Us at ETHGlobal New York

Ledger will be at ETHGlobal New York with a $10,000 developer prize pool split across five prizes.

We are looking for projects with real user value: agent payments, secure approvals, Ledger-backed identity, practical developer tooling, and workflows that make working with agents safer for the person who owns the assets.

ian c rogers,
Chief Human Agency Officer, Ledger

You might also like