How it works

You design the strategy.
AI agents execute it.

A team of autonomous AI agents researches markets, debates trades, manages risk, and executes on HyperLiquid - 24/7, fully transparent, adapting in real-time.

Launch a strategy
CoreX - Live
STREAMING

The flow

Three steps.
That's it.

No coding. Define what you want, deposit funds, and your agents start working.

01

Design your strategy

Pick pairs, set risk level, choose indicators, and tell the agents how aggressive to be. Or grab a pre-built strategy from the marketplace.

02

Deposit & activate

Bridge funds from Ethereum or Base, deposit into your strategy vault, and hit activate. Your AI agents spin up and start analyzing markets.

03

Watch them trade

Agents research, debate, and execute 24/7. Every decision streams live to your dashboard - you see what they're thinking and why.

The cycle

What happens every cycle

Your agents run a full analysis cycle autonomously. They decide when to run next - rechecking in minutes during volatile conditions, or waiting hours when the market is quiet. No fixed schedules. The agents adapt.

Gather intelligence

Live prices, technical indicators, social sentiment, and on-chain data - everything the agents need to understand the market right now.

Form a thesis

The Strategist analyzes the data and decides: buy, sell, or hold? It builds a trade plan with entry, stop-loss, and profit target.

Check the risk

The Risk Manager independently reviews every trade. Too risky? Too big? It can approve, adjust, or reject - no rubber stamps.

Execute the trade

Approved trades go to HyperLiquid with stop-losses, take-profit levels, and slippage guards - all placed automatically.

Monitor & protect

The Monitor watches positions 24/7 and can close early if the thesis breaks. It autonomously decides when the next cycle runs - rechecking in minutes during high volatility, or waiting longer when the market is calm.

Report everything

Every cycle produces a transparent summary. You see what was analyzed, what was decided, and why - in real time.

Under the hood

The nitty-gritty

Beyond the main cycle: timers, self-healing, and guardrails that keep your desk running 24/7.

Agent-controlled timers

In "agent" frequency mode, the Monitor decides when the next cycle runs. Volatile market? Recheck in 1–5 min. Quiet? Stretch to 1–2 hours. Minimum interval is configurable (e.g. 1 min). Open positions or big moves trigger a shorter fallback so the desk doesn't sit idle.

Watchdog

Every minute the scheduler checks all active strategies. If a strategy lost its timer (e.g. after a restart), it gets restarted using the configured cycle interval. No manual intervention - the system self-heals.

Evolution & auto-health

Every N cycles (e.g. 25), if the strategy has had no trades for a long streak, an LLM diagnoses why and can propose adjustments: cycle interval, prompt tweaks, signal exclusions. Applied changes create a new strategy version with a changelog.

Execution guards

Trade cooldown between trades; loss-streak cooldown (e.g. 3× normal cooldown after consecutive losses); max trades per day; drawdown limits. The Monitor is fed the current guard status so it doesn't infer from stale thread history.

LEVELS & MEMORY

The Monitor can output consensus levels (LEVELS: SUPPORT=… RESISTANCE=…) and a structured MEMORY block (thesis, key levels, patterns, warnings, lessons) so the next cycle has cross-cycle context - no amnesia between runs.

Your AI team

Three agents. One goal.

Specialized AI agents that challenge each other - so bad trades get caught before they happen.

🧠

The Strategist

Finds trading opportunities by analyzing charts, news, social media, and on-chain data. When it spots a good setup, it creates a detailed trade plan.

🛡️

The Risk Manager

Reviews every trade the Strategist proposes. Is it too risky? Too big? It can approve, shrink the size, or reject entirely.

👁️

The Monitor

Watches open positions 24/7. If the reason for a trade stops being true, it can close the position early. It also decides when the next cycle runs - recheck soon when volatile, or later when quiet.

How they work together

The Strategist wants to trade. The Risk Manager wants to protect. This tension means only high-quality trades get through.

Strategist proposes
Risk Manager reviews
APPROVE
or
REJECT
Execute on HyperLiquid

Your choice

You pick the model

Every strategy runs on the LLM you choose. We pull from 200+ models via OpenRouter - one API, every major provider. Pick budget, standard, or premium; the catalog is filtered for models that support tools and structured outputs so your agents can actually run.

One model per desk

When you configure a strategy or fund, you select the model that powers the Strategist, Risk Manager, and Monitor for that desk. Same brain for the whole team.

Tiers by cost

Budget for high-frequency cycles, standard for balance, premium when you want the strongest reasoning. Per-cycle cost is shown so you can tune for speed vs. quality.

Production-ready only

The picker only lists models that support tool calling and structured outputs, with enough context for full cycles. No guesswork - if it's in the list, it works with CoreX.

Model selection is in your dashboard when you create or edit a strategy, or when you configure a fund. Default is Gemini 2.5 Flash; switch to Mistral, Claude, GPT, or dozens of others in one click.

Intelligence

Where the data comes from

Your agents pull from multiple real-time sources - no guessing.

Live prices & charts

Real-time price feeds, order books, funding rates, and volume from HyperLiquid.

Technical indicators

45+ indicators - RSI, MACD, Bollinger Bands, Supertrend, Ichimoku, and more across all timeframes.

Social intelligence

Real-time monitoring of 20+ crypto Twitter accounts. AI scores relevance and impact.

Market events

Funding rate spikes, liquidation cascades, volatility shifts, open interest surges.

Web research

Agents search for macro news, catalysts, on-chain metrics - with source citations.

Technical edge

Patterns & indicators

Agents get a real-time snapshot of 45+ indicators and structured pattern signals across any timeframe - no manual chart reading.

get_market_contextbias + scorecalc_indicatorsany timeframe

Indicators

Computed in real-time for every timeframe. Agents reference these by name in their analysis - all available out of the box.

RSIMACDEMASMAHMAWMADEMATEMAVWMABBandsKeltnerDonchianATRNATRADXDMIStochRSIStochCCIMFIOBVCMFVWAPSupertrendIchimokuPSARPivot PointsFibonacciWilliams %RSqueezeVortexROCRVITSIAroonTRIXUltimate OscChaikin OscKSTCoppockElder RayMass IndexHeikin Ashi

Patterns & signals

Deterministic detectors and compiler tokens the Strategist can combine in natural language.

  • Volume - confirmation, spike detection, OBV divergence, participation score
  • Crosses - EMA/SMA crossovers, RSI threshold, MACD signal line
  • Consolidation - squeeze detection, BB width compression, Keltner/Donchian range
  • Trend - ADX strength, EMA ribbon, Supertrend, Ichimoku cloud position
  • Momentum - RSI/MACD/StochRSI alignment, ROC acceleration, TSI divergence
  • Reversion - VWAP stretch, BB band touch, RSI extremes, CMF divergence
  • Volatility - ATR regime (cool/hot), Keltner expansion, historical vs implied
  • Candlesticks - engulfing, hammer, doji, harami, kicker, shooting star, morning/evening star, three soldiers
  • Multi-timeframe - HTF alignment, divergence detection, VWAP anchored, indicator confluence scoring
  • On-chain - funding rate spikes, OI shifts, liquidation cascades, whale activity

Continuous improvement

How the agents learn

Cross-cycle memory, trade lessons, and auto-health evolution keep the desk from repeating the same mistakes.

MEMORY block

The Monitor outputs a structured MEMORY block every cycle: thesis, keyLevels, patterns, warnings, lessonsApplied. The next cycle receives this as context - no amnesia between runs.

--- AGENT MEMORY ---
thesis: "Long bias, watching 101k support"
keyLevels: "101200, 99800"
--- END MEMORY ---

Trade lessons

After each trade (or batch), the system can attach lessons: category, content, confidence. These are injected into the next Strategist prompt so the agent avoids repeating bad entries or ignores learned warnings.

Evolution & auto-health

Every N cycles (e.g. 25), if the strategy has had no trades for a long streak, the system runs a health check: an LLM diagnoses why (conditions too strict? wrong timeframes?) and can propose condition adjustments. Applied changes create a new strategy version with a changelog - the strategy literally evolves.

Cycle 1..NStuck?DiagnosePropose & applyNew version

Transparency

No black boxes. Ever.

You see everything your agents do - every thought, every decision, every trade.

  • Every agent conversation saved and viewable
  • Traceable sources - click through to verify
  • Live streaming of every cycle
  • On-chain proof of every trade on HyperLiquid
  • Clear cost breakdown per cycle

Under the hood

Built for production

CoreX isn't a wrapper around ChatGPT. It's a distributed system for real-time financial workloads.

Ready to try it?

Define your strategy, pick a risk level, and let AI agents trade HyperLiquid for you - around the clock.