Robinhood Chain Autonomous Trading Copilot is lets a user connect their wallet on the Robinhood Chain testnet, inspect holdings (TSLA, PLTR, AMZN, NFLX, AMD), and run AI agent to execute autonomously
Markets reward speed, context, and discipline. We’ve built a Robinhood Chain copilot that never sleeps: it ingests wallet balances and policy constraints, chats in natural language, and—in its fully autonomous mode—calls a PolicyVault contract to move tokenized equities on-chain. Users define max trade sizes, daily limits, and risk targets once; the agent handles the rest, logging every step with explorer links so compliance teams (or judges) can verify what happened.
Today’s state: the UI, policy vault, approvals tooling, and deterministic agent flow all work end-to-end on the Robinhood testnet. The copilot can explain your TSLA/PLTR positions, propose trades, and, when conditions are satisfied, execute via PolicyVault. We’ve also added demo switches for predictable judging: flip DEMO_AGENT_MODE on and the agent emits a scripted trade while the backend still returns real hashes. Limitations we’re still sanding down: the OpenAI planner occasionally declines to emit structured trades, and live MCP data feeds aren’t fully wired, so some suggestions fall back to deterministic logic. We’re fixing this by expanding the schema to accept richer GPT output and finishing the MCP adapters for Level 2 data.
Roadmap
V0 (shipped): Robinhood testnet equities, policy vault enforcement, explorer-backed audit trail, deterministic demo mode.
V1: Seamless GPT reasoning (switchable models), continuous MCP data, multi-token autonomy (not just TSLA).
Beyond: Multi-chain portfolio coverage, richer policy DSL (ESG, regional constraints), privacy rails, and optional on-prem agents for teams with proprietary data.
Our endgame: turn portfolio dashboards into collaborations with intelligent financial agents that monitor, explain, and trade faster than humans—while staying within the guardrails you set.
<ul><li><p><strong>On-chain executor:</strong> Designed and deployed PolicyVault.sol, an owner/operator contract that enforces per-token caps before moving Robinhood tokenized equities between treasury and user wallets. Added Hardhat scripts for deployment, policy configuration, and ERC‑20 approvals.</p></li><li><p><strong>Autonomous agent loop (in progress):</strong> Built a Next.js UI where users connect a wallet, set policy caps, chat with an AI copilot, review a suggested trade, and approve it. /api/agent supports deterministic (demo) and OpenAI-backed reasoning with stage/action logs. <strong>Limitation:</strong> the agent still stops short of executing autonomously in production mode—we can generate structured suggestions and hit the PolicyVault executor manually, but we’re debugging the final step so GPT-issued trades consistently pass policy checks and allowances.</p></li><li><p><strong>Execution pipeline:</strong> /api/trade routes manual or autonomous trades to the correct executor. We implemented demo, router, and PolicyVault executors; the PolicyVault path already returns real Robinhood testnet hashes once approvals/policies are in place.</p></li><li><p><strong>Explorer + telemetry:</strong> Added a Blockscout client so the agent logs live token balances, expanded the portfolio card to show all supported tokens, and log every policy save ([policy] updated … txHash). Each action/trade now links to the explorer for verification.</p></li><li><p><strong>Demo switches:</strong> Introduced DEMO_AGENT_MODE to produce deterministic agent responses while still exercising the on-chain pipeline, enabling reliable walkthroughs even while we finish the fully autonomous path.</p></li></ul><p>So far we’ve stitched together wallet connect → policy → agent reasoning → PolicyVault execution with audit trails; the remaining work is making the AI-driven execution path fully autonomous and data-rich.</p>
