A separate token per agent
Each agent authenticates with its own token and only the rights you grant. Revoke one at any time without touching the rest.
For AI agents
Agents do the work.You keep control.
Canonical lets AI agents prepare transactions, watch positions, and react to events — while keys, rules, and final approval stay with you.
The problem
An agent with your private key is a single prompt injection away from an empty wallet.
Giving an LLM signing power means one bad tool call, one poisoned webpage, or one hallucinated address can move real funds. But locking agents out entirely wastes what they're good at: watching state, preparing transactions, reacting fast. You shouldn't have to choose.
The Canonical answer
Agents propose. People approve. The wallet executes.
Your agent connects over MCP with its own token that physically cannot sign. When it wants to move funds, it files a proposal. The wallet decodes the proposal into plain English, checks it against your rules, and parks it in your inbox. Nothing leaves until the required people say yes.
Human approval is enforced
Mark an action as human-required and machine approvals can never satisfy it. The daemon enforces the rule at signing time.
Full audit trail
Every agent request, approval, and broadcast is written to a tamper-evident log on your machine.
Use cases
What this makes possible.
- A trading agent that drafts rebalancing transactions overnight, ready for your approval in the morning
- A payments bot that prepares payroll, where two founders must both sign off
- A monitoring agent that watches positions and proposes protective exits the moment conditions trigger