Try SAGE on Sepolia

A real wallet with two AI co-signers.

Connect your wallet and get your own 2-of-3 Safe, co-owned by you and two AI agents. Propose a payment and watch the policy engine decide whether the agents may sign it alone or must wait for you. It runs on the Sepolia test network, with no backend of ours.

Try it with your wallet Watch the 1-minute tour

Try it with your wallet

Your own Safe, two AI co-signers, five minutes.

Connect a wallet on Sepolia. You get a 2-of-3 Safe owned by you and two demo agents. The proposal agent proposes and signs first, you review, and then you sign or tell the signer agent to sign and send. Everything happens on the Sepolia test network, and your wallet pays the gas (about 0.003 test ETH).

Connect your walletMetaMask or any browser wallet, on Sepolia.
Create your SafeOwners: you, proposal agent …, signer agent …. Any 2 of 3.
Fund itSend it a little test ETH to play with. The signer agent also gets 0.0005 for gas, so it can send what you approve.
ETH
Ask the proposal agent to pay someone (or ask in the chat below, in plain language, for one or several payments)
To Amount (ETH)

What SAGE knows about this payment As shipped: nothing verified, so it should hold Pretend compliance and a price oracle are wired:

Value (USD) Business hours (Mon 11:00 BRT) Recipient seen before KYC verified AML clear AML flag
Review the queue The proposal agent signed each payment first. Check them. Then sign any one yourself with your wallet, or approve them all: the signer agent adds signature 2 and sends each one, in order, paying the gas. You can also just say it in the chat below: "sign them all, I've reviewed".

Activity

For this demo the two agents run in this browser tab with throwaway keys. In SAGE they run in your MCP server, and their keys stay in a vault. Your wallet is never asked for anything but a Sepolia transaction or a Safe signature, and you see each one before it goes.

Talk to your wallet

Bring your own AI. Ask it to pay someone.

Paste an API key from any provider below. Your browser talks to the provider directly, and the key never reaches a server of ours. The AI gets the same tools SAGE gives Claude over MCP, scoped to your demo Safe. It can propose and explain, but the policy engine decides, and only you can co-sign.

API key get one ↗ Model Remember the key on this device (otherwise it's forgotten when you close the tab)

Try asking

Hi. Connect your wallet and deploy your Safe above, then ask me to pay someone. I can propose and explain. The policy decides, and you sign.

Using the real SAGE with Claude

The demo above runs the agents in your browser. The real SAGE is an MCP server on your machine: its agents hold keys in an encrypted keystore or a vault, and Claude talks to it like this.

git clone https://github.com/criptografia-app/sage && cd sage && npm install
node setup.js            # agent keys → Safe owners → 2-of-3

Then add it to Claude Desktop (claude_desktop_config.json):

{ "mcpServers": { "sage": {
    "command": "node", "args": ["/path/to/sage/sage.js"],
    "env": { "SAGE_SAFE_ADDRESS": "0xYOUR_SAFE",
             "SAGE_PASSPHRASE_COMPLIANCE": "…", "SAGE_PASSPHRASE_EXECUTION": "…" } } } }

Then ask Claude things like "what's pending in the wallet?", "propose 500 USDC to Acme", "simulate 12k to a new address" or "signer, approve all I've checked". The tools are sage_info, sage_pending, sage_propose, sage_simulate, sage_approve, sage_approve_all and sage_history.

Policy engine, running in your browser

Would the agent sign this without you?

This is policyEngine.js from the repo, unchanged, with the example policy. No model is involved. Change the facts and see the decision.

…