> For the complete documentation index, see [llms.txt](https://docs.aspanfi.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.aspanfi.com/agent-payments/ai-capital-management.md).

# AI Capital Management

ASPAN's proposed AI manager coordinates the money an agent may spend with the money its owner has authorized for investment. It is intended to choose, adjust, and exit approved strategies against the owner's risk limits and liquidity needs.

**Status:** these are planned capabilities. The existing prototype does not demonstrate a production investment manager or verified improvements in returns.

### Start with when the money is needed

Upcoming payments, approved commitments, the owner's reserve floor, and withdrawal requirements determine which funds are eligible for investment. Unexpected spending and uncertain exits require a buffer; the highest advertised APY does not determine the allocation by itself.

### Compare eligible strategies

Within the approved universe, the manager should evaluate expected net return over the actual holding period, underlying exposure, concentration, capacity, and executable exit conditions. Both native ASPAN DeFi and planned ASPAN Vaults must meet these criteria.

Transaction fees, protocol charges, conversion costs, slippage, withdrawal costs, and time out of the market affect whether moving funds is worthwhile. Yield funded by temporary rewards should be distinguished from ongoing underlying income.

### Decide whether to hold, adjust, or exit

A change in spending plans, available liquidity, strategy conditions, or risk limits can justify a new allocation. Continuing to hold can be the better decision when the expected benefit of switching does not cover its costs.

For example, funds needed for a near-term service payment should not enter a strategy whose exit conditions conflict with that deadline. A nominally higher yield does not solve a liquidity mismatch.

### Execute inside the mandate

Independent permission checks must enforce allowed protocols, assets, exposure, and spending limits. The AI cannot authorize its own expansion of scope. Missing or unreliable inputs should trigger a defined fallback, such as retaining liquid funds or requesting owner review, rather than an unbounded action.

Withdrawal planning must track the asset and amount actually received. Investment NAV, an exit request, and confirmed payment funds are different states.

### Explain and evaluate decisions

The intended record connects the inputs and policy used, the decision and its reason, expected costs and timing, actual execution, and the subsequent result. This enables evaluation of net returns, exit delays, liquidity shortfalls, and manual interventions.

Initial evaluation should compare AI management with simple fixed-reserve and periodic-rebalancing rules using the same assets and constraints. Simulated alternatives are estimates, not realized outcomes. Improvements should be tested before expanding real-fund authority; users' money should not finance unrestricted model exploration.

Returns come from underlying assets and protocols. AI is intended to improve decisions and coordination, and does not guarantee low risk, the highest yield, or uninterrupted liquidity.
