How Mirae Works

Mirae separates an agent-assisted action into six stages. Each stage has a defined output, and execution never skips ahead of approval.

1. 1

   Prompt

   The user types what they want to do.

2. 2

   Understand

   Mirae identifies the intended action, constraints, and missing information.

3. 3

   Discover

   Mirae checks supported protocols, data sources, balances, and market conditions.

4. 4

   Plan

   Mirae produces a human-readable execution plan and transaction preview.

5. 5

   Approve

   The user reviews permissions, costs, and risks before signing.

6. 6

   Execute

   The approved transaction is submitted to Solana Chain and recorded with an onchain receipt.

Where the trust boundary sits

The agent proposes. The policy layer validates. The wallet authorizes. The network settles. No stage may substitute for another, and a failed policy check stops execution.

What a plan contains

• Network and destination contract
• Assets, amounts, and expected output
• Estimated network fees
• Slippage or other execution constraints
• Required token approvals
• Material risks and conditions that would invalidate the plan