Private preview · Agent-managed capital

A governed way to delegate financial operations to agents.

AgencyAI Capital Control extends our agentic-commerce control plane into capital management. A human defines the mandate; specialized agents observe, explain, propose, assess risk, enforce policy, execute within bounded authority, and reconcile what happened.

The missing layer is not more data. It is controlled delegation.

Dashboards show positions. Trading terminals submit transactions. DeFi protocols provide primitives. Capital Control connects those pieces into one understandable operating loop so a human can delegate attention and bounded actions without surrendering visibility or accountability.

Observe → Explain → Propose → Risk-check → Policy-check → Approve or block → Execute → Reconcile → Measure

For the human

One clear next action

See what changed, what the agents recommend, what could go wrong, and whether anything requires approval.

For the system

Typed, auditable decisions

Mandates, observations, evidence, risk decisions, approvals, executions, and reconciliations are structured records—not loose agent messages.

For capital

Bounded authority

Policies define assets, venues, limits, liquidity, slippage, expiry, approval thresholds, and prohibited actions before execution is possible.

Six roles. One control plane.

The agents are specialized by responsibility. The policy and audit layers remain deterministic and independent from the model that proposes an action.

Portfolio ObserverBalances, positions, exposure, liquidity, P&L, and source health.
Research AgentMarket, protocol, rate, governance, and external evidence with provenance.
Strategy AgentStructured DCA, rebalancing, cash-management, or allocation proposals.
Risk AgentLiquidity, volatility, oracle, counterparty, smart-contract, and drawdown analysis.
Policy & AuthorizationDeterministic allowlists, limits, expiry, approval tiers, and fail-closed decisions.
Execution & ReconciliationRestricted execution, receipts, expected-versus-actual accounting, and audit.

Built from the AgencyAI systems already in motion.

Agentic Commerce Control

The authority and evidence layer

Identity, delegated authority, policy, approval, adapters, receipts, reconciliation, and trust patterns extend naturally from commerce into capital operations.

TradeDesk

The human oversight surface

Plan, Checklist, Positions, Journal, Market, paper trades, and plain-language review become the user experience for mandates and proposals.

x402 and CyberAgency

Paid intelligence before action

Agents can purchase risk assessments, market data, or protocol research as evidence before a proposal reaches policy review.

DeFi and RWA adapters

Execution venues, not the product

Stablecoins, lending markets, tokenized cash, and future RWA providers can plug into the same control loop without defining the user experience.

A calm experience over sophisticated machinery.

Capital Control is designed as a private workbench, not a protocol console. Users work with capital sleeves and decisions; technical details remain available progressively when needed.

1Today

What is my capital doing? What changed? Is anything urgent?

2Proposal

What is proposed, why now, what evidence supports it, and what is at risk?

3Decision

Approve once, approve a rule, reject, request more analysis, or pause the strategy.

4Proof

See what was authorized, what executed, what reconciled, and what was learned.

What is working today

The product is in private preview. The evidence is operational, not a claim of investment performance.

Control loopPhases 0–6 implemented
IntelligenceTradeDesk read-only adapter and paper replay
Safety53 Capital and connector tests passed
ExecutionOne bounded Base Sepolia transfer reconciled
Important boundaryCapital Control is not an autonomous fund, broker, or promise of alpha. Mainnet, live capital, unrestricted DeFi, leverage, bridges, and RWA execution remain outside the current preview.

Where this can go

The next strategy should be deliberately boring: a small, isolated stablecoin treasury sleeve with human approval, established venues, no leverage, and complete measurement after fees and execution costs.

Next

Persist the full control loop

Exercise reservation, simulation, connector acknowledgement, receipt, reconciliation, replay rejection, pause, and recovery as one durable workflow.

Then

Controlled treasury strategies

Test stablecoin cash management, rule-based DCA, rebalancing, and protocol-risk surveillance in isolated sleeves.

Later

Tokenized financial assets

Extend the same mandates and evidence model to tokenized cash, RWA eligibility, and institutional treasury operations.

Interested in the control plane for agent-managed capital?

We are developing this as a private, human-supervised system for people and organizations that want to delegate financial operations without delegating responsibility.