How Strat AI Uses AI — Public AI Transparency & Governance Disclosure
Learn exactly what AI models do in Strat AI, structural limits, models and routers used, input data boundaries, and known limitations.
How Strat AI Uses Artificial Intelligence
We use large language models to help analyze crypto trade setups. Since this capability is easily misunderstood, here is exactly what the AI does, what it is given, where its boundaries lie, and where it fails.
1. What the AI does
The AI evaluates one crypto pair at a time and argues about whether a trade setup is worth taking. Concretely, when you run an analysis the model:
Fetches Market Data
Calls quantitative tools to retrieve real market data: candles, deterministic indicators, volume profile, derivatives positioning metrics (funding, open interest, liquidations), order-flow proxies, and recent news.
Identifies Data Gaps
Reads the returned values and notes where data is missing or incomplete, explicitly reporting unavailable feeds.
Reasons in the Open
Every tool call, argument passed, and numeric value returned is streamed live to your screen as it happens.
Proposes Setup or Declines
Either proposes a bracket (direction, entry, stop loss, take profit, holding period) or declines to recommend.
Adversarial Self-Critique
A separate Bear Agent pass attacks the proposal. Objections are included directly in the final output rather than discarded.
Key Distinction: The output you receive is research — an opinion about an instrument with reasoning attached — not an execution command or personal advice.
2. What the AI is not allowed to do
These are hard structural constraints enforced by code, not soft policies:
It cannot place, modify, or cancel an order
No capability to reach an exchange order system exists anywhere in the product. The internal exchange interface uses read-only API keys and contains no trading or withdrawal methods. Strat AI provides analysis and risk tooling, not trade execution.
It cannot give you personal financial advice
Strat AI does not tailor recommendations to your capital, income, net worth, portfolio, or position size. If asked a personalized question ("how much of my capital should I put in this?"), a deterministic pre-LLM validator automatically declines and redirects to impersonal analysis of the instrument.
It cannot override hard risk validation rules
A proposed trade whose stop loss is closer than 1.5× the instrument's 14-period ATR, or whose reward-to-risk ratio falls below the profile floor (1:1.3 intraday / 1:2 swing, positional, and perpetuals), is automatically rejected by a deterministic validator regardless of the model's confidence.
3. What the AI is given as input
| Input Data | Source & Processing |
|---|---|
| Price & Volume History | Exchange WebSocket & REST APIs (spot and perpetuals), stored in local time-series database. |
| Live Ticks | Exchange WebSocket streams (spot and perpetuals), 24/7. |
| Computed Indicators | Deterministic pure functions (RSI, MACD, EMA, ATR, ADX, VWAP, Volume Profile). Not AI-generated. |
| Derivatives Positioning Data | Funding rates, open interest, liquidation clusters, long/short ratio, and perpetual basis vs spot. |
| Recent News Headlines | Third-party financial news feeds classified for market sentiment. |
4. Which models we use
Models are configurable per deployment. The default models and routing endpoints are:
| Role | Default Model | Router Endpoint |
|---|---|---|
| Main Research Loop (FIND) | openai/gpt-4o | OpenRouter |
| DEBATE Roles (Bull / Bear / Judge) | openai/gpt-4o / gemini-2.5-flash | OpenRouter |
| Market Anomaly Commentary | deepseek-ai/DeepSeek-V3-0324 | freemodel.dev |
| News Sentiment Scoring | deepseek-ai/DeepSeek-V3-0324 | freemodel.dev |
Router Disclosure: Both OpenRouter and freemodel.dev are OpenAI-compatible routers rather than the original labs running the models. They forward requests to inference hosts. We do not train these base models; our proprietary intellectual property is in the quantitative tooling, prompt composition, deterministic risk verification, and guardrail engines.
5. Limitations & Probabilistic Nature
Language model reasoning is probabilistic. It is not a prediction of the future, and it carries no performance guarantees:
- •Confident errors: Language models can sound articulate while drawing incorrect conclusions. Confidence in writing carries zero correlation with accuracy.
- •Scope bounded to provided data: The model only evaluates market indicators and headlines passed to it. It has no visibility into upcoming token unlocks, exchange incidents, or unannounced policy shifts.
- •Past patterns do not guarantee outcomes: Market regimes shift. Historical setups do not guarantee future price action.
- •Missing data is reported, not guessed: If open interest, funding, or news feeds are offline, the system marks them unavailable rather than substituting synthetic numbers.
Regulatory & Risk Notice
Strat AI is a quantitative market research and decision support terminal. It does not execute trades, manage funds, or offer personalized investment advice, and is not a registered investment adviser. Crypto assets are highly volatile, and trading spot pairs or leveraged perpetual futures involves substantial risk of loss, including liquidation of your entire margin. Past performance is not indicative of future returns.