Skip to main content

AI Stack

Alphio AI combines a proprietary quant model, a self-developed agentic workflow technology architecture, LLMs, and a wallet-based on-chain execution layer for research, analysis, and execution.

1. Proprietary quant model​

At the core of Alphio AI is a proprietary quant model built for finance-first decision support. It focuses on turning noisy market inputs into structured signals and comparable outputs that downstream workflows can use reliably.

It is used for tasks such as:

  • Signal extraction - converting market and fundamental inputs into machine-usable features. The signal models are trained primarily on US markets, so coverage of other markets (for example Hong Kong stocks) is thinner.
  • Ranking and scoring - comparing assets, setups, and event candidates across a common framework.
  • Financial interpretation - supporting workflows that need domain-specific judgment rather than generic text generation.

This layer powers much of the structured intelligence behind research surfaces, market monitoring, Signals, and agent-assisted trading workflows.

2. Agentic workflow technology architecture​

Alphio AI is not a single model endpoint. It runs on a self-developed agentic workflow technology architecture that breaks complex tasks into steps, routes each step to the right capability, and manages the handoff between data, reasoning, and action.

A typical workflow can include:

  • Intent understanding - identifying what the user is asking for and which workflow to run.
  • Data assembly - pulling the right market, news, account, or on-chain context for the task.
  • Multi-step reasoning - coordinating structured analysis, follow-up checks, and exception handling.
  • Action gating - separating suggestion from execution, with approval checkpoints where required.
  • Ongoing monitoring - supporting alerts, recurring checks, and task-style follow-through.

This architecture is what allows Alphio AI to act more like a finance-native agent system instead of a single chat completion.

3. LLM layer​

LLMs are one layer of the stack, used where natural-language understanding and generation are the best interface.

They are primarily used for:

  • Natural-language interaction - understanding user requests and follow-up questions.
  • Answer composition - turning structured outputs into readable summaries, explanations, and recommendations.
  • Multi-modal tasks - handling screenshots, charts, documents, and other unstructured inputs where applicable.
  • Flexible general reasoning - covering open-ended questions that are broader than a narrow quant pipeline.

LLMs improve usability and breadth, but they are not the sole source of Alphio's financial intelligence. In practice, they work together with the proprietary quant model and the agentic workflow architecture.

4. On-Chain Execution & Blockchain Integration​

Alphio does not take custody of your funds: on-chain assets stay in wallets that belong to your account, and stock orders go to your own broker account.

Wallets​

  • Alphio wallet - Alphio creates embedded wallets for your account through Privy — one EVM wallet and one Solana wallet — when you first need them, for example when you first open Deposit or trade. These are the wallets used for on-chain trading and deposits.
  • External wallets - New external wallets (such as MetaMask or WalletConnect) can no longer be added. External wallets linked before remain visible so you can withdraw from them.

Execution venues​

  • Hyperliquid perpetuals (deposits via Arbitrum).
  • Polymarket prediction markets (orders signed on Polygon, settled in USDC).
  • On-chain tokens from Crypto Trade → Tokens — Solana meme coins, plus tokens on Ethereum, Base, BNB Chain, and Monad. See Meme coins.
  • US stocks are not on-chain; orders route to your connected broker (via SnapTrade or Robinhood).

How orders are signed​

Not every order pops up a wallet confirmation. Depending on the feature:

  • Hyperliquid orders are signed by a Hyperliquid API ("agent") key that your wallet authorizes. Alphio stores this key encrypted and uses it to place orders on your behalf; Hyperliquid API keys can trade but cannot withdraw funds.
  • Delegated signing - for fast meme-token trading (which you turn on yourself), Polymarket orders and the Predictions withdrawals you confirm with an email code, and automatic Hyperliquid deposits and Hyperliquid trading-access approvals, a server-side signer is added to your embedded wallet through Privy. Each signer is restricted by a policy — for example, fast trading has per-trade and 24-hour limits.
  • Selling EVM tokens needs a one-time approval per token on each chain, which you agree to in the first-sell setup (Alphio pays the network fee). It also delegates your EVM wallet on that chain to the Calibur smart-wallet contract (EIP-7702), and that delegation stays in effect. Sells skip the wallet confirmation only if you also turn on fast selling, and then only within your fast-trading limits.

Real-time on-chain data​

Separately from execution, the agentic workflow architecture reads on-chain data across ecosystems — Ethereum, Solana, Base, BSC, Arbitrum, Polygon and others — via providers such as DefiLlama and CoinGecko Onchain / GeckoTerminal. This is research data only; broad read coverage does not mean Alphio can trade on every chain it reads. Signals include:

  • large wallet movements (smart money flows)
  • liquidity changes and pool depth
  • token launches and early trading activity
  • funding rates and open interest for perpetuals

Verifying trades​

  • Hyperliquid trades can be checked on Hyperliquid's own explorer (Hyperliquid is its own chain, so they don't appear on Arbiscan).
  • Polymarket transactions are on Polygonscan.
  • On-chain tokens are on Solscan (Solana), Etherscan, BaseScan, BscScan, or MonadScan.

Security-first design​

Agents operate under permission boundaries: they can only perform actions you have authorized through workflow settings, delegated-signing policies, or one-time approvals. See Wallet and trade security for more.

5. Privacy and execution boundaries​

What stays on Alphio infrastructure vs what leaves to a provider:

  • On Alphio infrastructure - your account, broker / wallet connections, the encrypted Hyperliquid API key, order history, watchlist, position state, alert configurations, Credits balance and ledger. None of this is sent to external model providers as raw credentials.
  • Sent to model providers when needed - the textual content of your query and the prompt context Alphio assembles for that workflow. Provider-side handling follows the applicable commercial and privacy terms.
  • Never sent - payment instrument details, private keys, and raw broker credentials.

When you ask Alphio AI to trade, the model drafts the order intent; Alphio's own services validate it and execute it on the relevant rail, after any required confirmation.

For AI agents and tools

This page is available as plain Markdown at https://alphio.ai/docs/technology/ai-stack.md. An index of every page, with a one-line summary each, is at https://alphio.ai/docs/llms.txt.