Catch Investments Beta Sign in

Transparent by design

How it works

Catch combines AI research, deterministic guardrails, and human approval to manage a connected Robinhood account. AI can recommend an action, but it cannot approve or submit a trade by itself.

Live brokerage with human controls
01
Research layerAI researches

Portfolio context, market research, thesis, and risks.

02
Safety layerCode validates

Trusted data and every deterministic guardrail.

03
Human layerYou approve

A fresh passkey authorizes the exact selection.

  1. 01

    Start safely

    Start a review

    A manual request or scheduled review starts the same workflow. It first confirms:

    • Your Catch account is active.
    • Robinhood is connected and verified.
    • Global, administrator, and user trading controls allow the requested action.
    • Market data and AI budgets are available.

    If a prerequisite fails, the workflow stops safely and records the reason.

  2. 02

    Authoritative context

    Capture the Robinhood snapshot

    The system retrieves the current account state and gives the research model only the context it needs:

    Cash, buying power, and total equityOpen positions and cost basesTrusted current pricesRealized and unrealized P&LStrategy assignments and allocation targetsEvery active trading guardrailExcluded industries and ETF-only preferences

    Refresh data can capture a new snapshot without using AI. AI reviews and proposals build on that stored snapshot.

  3. 03

    Research proposal

    AI recommends an action

    The research model can return:

    BUYInvest the specified dollar amount.
    SELLSell the specified dollar amount.
    REBALANCEMove a position toward a desired value.
    HOLDRetain an existing position.
    NO ACTIONDo nothing.

    Each recommendation includes a ticker, strategy, confidence score, thesis, risks, and suggested dollar value. The AI does not choose execution prices, share quantities, fills, or final guardrail decisions.

  4. 04

    Independent prices

    Obtain trusted market data

    The system independently retrieves security classification and current quotes from the configured market-data provider. The model is not allowed to provide:

    • Execution prices
    • Share quantities
    • Fill results
    • Guardrail decisions

    Missing, stale, or inconsistent provider data causes the action to be rejected.

  5. 05

    Deterministic safety

    Validate the proposal independently

    Before anything can be approved or sent to Robinhood, the risk engine checks:

    Supported U.S. stock or ETFRegular market sessionQuote freshness and consistencyRobinhood connection and trading switchesETF-only and excluded-industry rulesMarket-cap limitsDaily trade and initial-purchase limitsDaily-loss limits and the hard 15% backstopMaximum 15% of equity per tradeMaximum 20% projected exposure per tickerMinimum cash reserve30-day turnoverStrategy driftMinimum and maximum trade valuesAvailable cash or sharesMinimum holding period for sales
    Confidence thresholds90% Passive75% Balanced70% Aggressive

    Any failed check rejects the proposal and records the reasons.

  6. 06

    Human-controlled execution

    Approve the exact orders

    Proposals appear in the Approvals inbox. You choose which orders to authorize, then confirm that exact selection with a fresh passkey. Approval expires after five minutes and is bound to the selected orders.

    APPROVALFresh passkey
    RECHECKQuotes + risk
    SUBMISSIONRobinhood only
    Control

    Catch never treats an AI recommendation as permission to trade. A human approval and a final execution recheck are required.

  7. 07

    Auditable accounting

    Submit, reconcile, and record

    After approval, the execution workflow submits the authorized orders to Robinhood and handles recoverable or uncertain outcomes. The system records:

    Trade intent and approval selectionRobinhood order referenceSubmission and fill statusPosition quantity and cost basisRealized or unrealized P&LFresh portfolio snapshotsActivity and failure explanations

    Retries and recovery paths are designed to avoid duplicate orders and to fail closed when the outcome is uncertain.

The important distinction

AI finds the opportunity. Code and people control the action.

The AI provides research and recommendations. Deterministic code decides whether an action is safe, and you authorize the exact order with a passkey before Robinhood can receive it. HOLD, rejection, and no-action are all normal successful outcomes.