Skip to content

01Product

Four modules. Two execution paths. One place the risk lives.

Build strategies in the Anaconda stack. Route listed equities through Robinhood's agentic account. Route tokenized assets and DeFi through Robinhood Chain. Keep keys, budgets and research reproducible.

  1. 01

    Workbench

    A Miniconda environment with Python 3.12, pandas, numpy, JupyterLab, web3 and httpx.

    The research half. A pinned environment file is the whole contract: clone it, activate it, and a strategy written today reruns identically next quarter. Notebooks stay notebooks. Nothing is hidden behind a hosted runtime.

    • Python 3.12 from conda-forge
    • pandas, numpy, JupyterLab
    • web3 + httpx via pip
    • Any MCP-capable client
  2. 02

    Broker Bridge

    Connects an agent to the official Robinhood MCP, inside a user-funded agentic sub-account.

    The listed-equity half. The agent never touches a primary account and never sees a password. It speaks to the official MCP surface, scoped to a sub-account you funded on purpose. Every order fires a push notification. One kill switch stops all of it.

    • Official Robinhood MCP only
    • Dedicated agentic sub-account
    • Push notification per order
    • Kill switch halts routing
  3. 03

    Chain Bridge

    An EVM client pointed at Robinhood Chain. Stock Tokens, ERC-20s, swaps, lending. ETH gas.

    The 24/7 half. Same notebook, different rail: an EVM JSON-RPC client aimed at Robinhood Chain, an Arbitrum Orbit L2. Tokenized stocks and stablecoins settle on-chain and keep trading when the listed market is shut.

    • RPC and chain ID from the official docs
    • Stock Tokens and ERC-20s
    • Swap and lend legs
    • Agent wallet; keys stay yours
  4. 04

    Guardrails

    Max notional, allowlists, daily loss cap, human approval on size, separate keys per rail.

    The half that matters. Limits live in a YAML file next to the strategy, versioned with it, and they are checked before anything is routed. Broker credentials and chain keys never share a scope, so a compromise on one rail does not become a compromise on both.

    • Per-order max notional
    • Ticker and token allowlists
    • Daily loss cap, hard stop
    • Human approval over a threshold

05Status surface

What the desk tells you without being asked.

ah-env · sessionpaper
  • a

    Connection state per rail

    Broker and chain report separately. If one rail is down the agent is told, and a strategy that needs both does not run.

  • b

    Mode is always visible

    Paper is the default and it is printed on every surface. Going live is a deliberate edit, never a side effect of a config change.

  • c

    One key stops everything

    Shift + K flips the kill switch: open orders are cancelled and new routes are refused on both rails.

The sequence from a clean machine to a logged fill, end to end.

How it works