ORC-1 · Private beta on Robinhood Chain

Trading, automated by
machine intelligence.

ORC is an applied AI lab building an autonomous trading program for Robinhood Chain. It reads the market, the chain, and the crowd, then executes with discipline no human trader can hold for 24 hours a day.

24 / 7Autonomous execution
< 400 msSignal to order
3 enginesTrade · Deploy · Scan
EVMRobinhood Chain native
model ORC-1 · 72B
inference H100 cluster
latency 312 ms
status online
Built on
Robinhood ChainArbitrum OrbitEthereumPyTorchNVIDIA H100CUDA
Research programs

One program.
Three engines.

Most trading bots follow rules. ORC learns. Each engine is a model trained on its own domain, sharing one world state so a rumour on X becomes a position on-chain within seconds.

01 / TRADE

Trading Engine

A reinforcement-learned execution policy trades coins on Robinhood Chain. It sizes, enters, and exits positions from a live model of liquidity, momentum, and risk.

RL · order-flow · MEV-aware routing
02 / DEPLOY

Token Deployment

Describe a token and ORC writes, audits, deploys, and bundles the launch. Contract, liquidity, and initial distribution are handled as one atomic operation.

Contract synthesis · bundling · launch ops
03 / SCAN

Social Intelligence

Language models read X, Telegram, Discord, and Farcaster in real time, separating narrative from noise and scoring attention before it becomes price.

NLP · attention scoring · bot filtering
04 / COMPUTE

Training Infrastructure

ORC models are trained on dedicated H100 clusters against years of chain and market data, then continuously fine-tuned on live outcomes.

H100 · continuous fine-tuning · eval harness
System

From signal to settlement.

Every trade passes the same pipeline. Perception, reasoning, risk, execution, and review, with a hard risk layer no model can override.

01PERCEIVE

Ingest

Chain state, order books, mempool, and social streams merged into one tensor of the world.

02REASON

Predict

ORC-1 estimates short-horizon return distributions and the probability the crowd is early or late.

03GUARD

Risk

Position limits, drawdown stops, and contract safety checks enforced outside the model.

04ACT

Execute

Routes through Robinhood Chain with private submission to avoid front-running and slippage.

05LEARN

Review

Every fill is scored against the prediction and folded back into training.

orc-1 · live session · robinhood-chain

    
Models

Trained to trade, not to chat.

General models know a lot about everything. ORC models know one thing deeply: how price, liquidity, and attention move together on-chain.

ORC-1 FOUNDATION · 72B
Serving

The core reasoning model. A transformer trained on tokenised order flow, chain events, and social text, aligned to a single objective: risk-adjusted return over a rolling window.

72BParameters
1.4TTraining tokens
128kContext
ORC-Exec POLICY · 3B
Serving

A small, fast execution policy distilled from ORC-1. Runs at the edge next to the sequencer and decides how, not whether, to trade.

3BParameters
38 msMedian inference
RLTraining method

ORC-Scan Serving

Multilingual social encoder that scores posts for intent, credibility, and expected reach.

ORC-Forge Serving

Code model for contract synthesis and audit. Generates, tests, and deploys ERC-20 launches end to end.

ORC-2 In training

Next-generation foundation model. Multi-chain, longer horizon, native portfolio reasoning.

Eval harness Open

Replayable backtests on real chain history so every model change is measured before it trades.

Robinhood Chain

Built for the chain
retail actually uses.

Robinhood Chain is an EVM network designed to bring traditional retail traders on-chain. ORC is native to it: same wallets, same assets, same rails, with a model doing the work.

EVM-compatible: standard wallets, contracts, and tooling.
Non-custodial. ORC signs with delegated, scoped permissions you can revoke at any time.
Private order submission to protect against sandwich attacks.
Every action is on-chain and auditable.
EVMExecution layer
~250 msBlock time
Sub-centMedian gas
24/7Markets
Safety

The model proposes.
The guard disposes.

Autonomy without limits is just a faster way to lose money. ORC keeps a deterministic risk layer between every model output and the chain.

Limits

Hard position and drawdown caps

Per-asset, per-day, and per-portfolio limits you set. The model cannot raise them.

Scoped keys

Delegated, revocable permissions

ORC never holds your seed. Session keys are scoped to approved contracts and expire on schedule.

Audit

Every decision explained

Each trade ships with the signals, the prediction, and the risk check that let it through.

Early access

Stop trading.
Start directing.

ORC is onboarding a limited cohort of traders on Robinhood Chain. Join the list and we will reach out when your seat is ready.

No spam. Cohort invites only. Trading involves risk of loss.