Agentic Wallets on TON
Self-custody wallets that leverage AI agents to execute transfers, swaps, payments and DeFi on TON

Agentic Wallets empower AI agents with autonomous and secure access to funds and services
AI agents are becoming power users on digital platforms by accessing APIs and DeFi protocols to coordinate with other agents – but require manual approval for each transaction.

Workflow
Create
The agent generates keys and deploys a wallet contract with your address set as the owner
Fund
You transfer funds to the agent wallet and define its available balance to spend
Operate
The agent autonomously signs and executes transactions using its own key
Security and control
It’s time to give agents their own wallet
Common questions
The agent can only spend what you sent to its wallet. Your main wallet is completely separate. However, if the agent behaves incorrectly or executes harmful transactions, it can spend the entire balance of its wallet before you react. The risk is capped to the funded amount — but that amount is still real money. Fund agents with what you're comfortable losing.
Tap on "Revoke" and confirm the transaction. The operator is deactivated and you can safely withdraw remaining funds.
Your master key is never shared with agents. The agent has its own operator key that can only spend from its own wallet. All logic is on-chain, open source, and verifiable. No servers, no middlemen.
As many as you want. Each agent gets its own wallet with its own balance. You see and control all of them from one dashboard.
Install skills via npx skills add ton-connect/kit/packages/mcp and ask your AI-agent to "create agentic wallet". The agent handles key generation, deployment, and onboarding. Skills are compatible with any AI client (Cursor, Claude Desktop, Windsurf, etc.). See the GitHub repo for full docs.
Not yet. This is a developer preview. The contracts are open source and available for review on GitHub, but a formal audit has not been completed. We recommend using testnet for experimentation until the audit is done.