Core Concepts
Key terminology and concepts for understanding PerpDesk.
Perpetual Futures Primer
Perpetual futures (perps) are derivative contracts that let traders take leveraged long or short positions on an asset without an expiration date. Unlike traditional futures, perps use a funding rate mechanism to keep the contract price aligned with the spot price.
Key characteristics:
- No expiry — Positions can be held indefinitely
- Leverage — Typically 1x to 125x depending on exchange and asset
- Funding rates — Periodic payments between longs and shorts to maintain price parity
- Mark price — Used for liquidation calculations, derived from spot and futures prices
- 24/7 operation — No market close, no circuit breakers, continuous global participation
PerpDesk is built specifically for these dynamics.
Prediction Markets Primer
Prediction markets are venues where users trade contracts on real-world outcomes — elections, economic releases, sports, and event-driven questions. Each contract has a YES side and a NO side, both priced between 0 and 1.
Key characteristics:
- YES / NO contracts — Buy YES if you think the outcome will happen, NO if you don't
- Mid-price ≈ implied probability — A YES quoted at 0.62 reflects the market's ~62% probability the event resolves true
- Resolution — At settlement, the winning side pays $1 per contract; the losing side pays $0
- Two flagship venues — Polymarket (crypto-native, on-chain) and Kalshi (US-regulated, CFTC-designated)
- Edge — The gap between your calibrated probability and the market mid
PerpDesk's prediction-market analyst, Plop, reads order books, base rates, and quote drift on these venues and publishes a calibrated p(YES) alongside the gap to the market.
Market Regimes
A market regime describes the current behavioral state of a market. Vanta (the regime detection agent) continuously classifies markets into five states:
Trending
Characteristics: Strong directional momentum with sustained movement
- Price making higher highs/lows (uptrend) or lower lows/highs (downtrend)
- Volume confirming the direction
- Support/resistance levels shifting in the trend direction
Best strategies: Momentum, breakout, trend-following. Worst strategies: Mean-reversion, range-bound.
Ranging
Characteristics: Price oscillating within defined boundaries with no directional bias
- Bouncing between support and resistance
- Volume declining as traders lose conviction
Best strategies: Mean-reversion, oscillator-based. Worst strategies: Momentum, breakout.
Volatile
Characteristics: Rapid price swings with unclear direction
- High intra-day ranges
- Whipsaws that catch stop-losses
- Liquidation cascades possible
- Often triggered by macro events or funding rate spikes
Best strategies: Short-term scalping, volatility-adjusted positioning. Worst strategies: Large positions, trend-following.
Quiet
Characteristics: Low volatility, narrow ranges, often preceding a significant move
- Compression pattern with low volume
- Can last hours or days before breakout
Best strategies: Waiting/preparation, breakout-ready positions. Worst strategies: Forcing trades.
Transitioning
Characteristics: Market shifting from one regime to another
- Regime signals conflicting
- Uncertainty elevated, high risk of whipsaw
Best strategies: Reduced sizing, flexible exits, wait for confirmation. Worst strategies: Aggressive entry, heavy leverage.
Why regimes matter: A momentum strategy that works in trending markets will lose money in a range. Riven adapts its consensus recommendations based on the regime Vanta detects.
Agent Pipeline
PerpDesk's 7 agents follow a structured pipeline. Six form the crypto perps consensus chain; the seventh — Plop — runs a parallel pipeline for prediction markets.
Crypto Perps Consensus Chain
In the perps chain, each agent's output feeds the next:
Step 1: Analyze (parallel)
- Vanta detects the market regime using technical indicators (EMA, RSI, MACD, Bollinger Bands, ATR, StochRSI)
- Luma aggregates cross-venue signals (funding rates, open interest, volume anomalies, liquidation data)
- Meridian monitors macro context (economic events, regulatory news, cross-asset signals)
These three agents run in parallel on every candle close.
Step 2: Consensus
- Riven receives input from Vanta, Luma, and Meridian
- Riven runs a consensus debate, weighing and synthesizing inputs
- Produces one broadcast recommendation per symbol: entry, targets, stops, position sizing
Step 3: Risk Filter
- Brix reviews the recommendation against your personal risk parameters
- Brix checks: position sizing limits, total exposure, leverage caps, daily loss thresholds
- Brix is a deterministic rules engine — it enforces hard limits with no AI interpretation
Step 4: Execution Guidance
- Koda receives the risk-filtered recommendation
- Koda adds execution guidance: entry style (sniper, ladder, or market), timing considerations
- The final package is presented to you for approval
Once you approve, the trade executes on your connected exchange with continuous stop-loss and take-profit monitoring.
Coming Soon: Lucid (performance tracking), Mentor (education), and Assistant (custom data) are in development and will extend the pipeline with trade analytics, learning insights, and custom data integration.
An Orchestrator coordinates multi-agent communication behind the scenes, routing messages between agents and managing the pipeline flow.
Prediction-Market Pipeline (Plop)
Plop runs a separate, simpler pipeline for prediction-market contracts. It does not flow through Riven, Brix, or Koda.
- You drop a Polymarket or Kalshi market URL, slug, or contract ID into chat — or Plop surfaces one from the discovery feed.
- Plop pulls the order book, quote drift, historical base rates, and surrounding news.
- Plop publishes a calibrated p(YES), the market mid, the edge, a confidence score, and citations.
Today's reads are advisory. Signed-order execution into Polymarket and Kalshi is on Plop's roadmap.
Autonomy Levels
Every PerpDesk agent has an autonomy level you control:
Monitoring Mode
Agent observes and logs insights, takes no action. Best for:
- Learning the system
- Understanding agent logic before trusting execution
- Reviewing recommendations before they happen
Advisory Mode
Agent proposes actions and waits for your approval. Best for:
- Trades where you want final say
- Building trust with the system
- Most traders' default setting
Active Mode
Agent acts within your configured boundaries. Best for:
- Routine execution of approved recommendations
- Risk management (Brix always operates at this level minimum)
- Continuous monitoring during off-hours
You can set different autonomy levels for different agents. For example:
- Vanta, Luma, Meridian: Active (analysis agents run continuously)
- Riven: Advisory (you review strategy recommendations)
- Brix: Always active (risk enforcement is non-negotiable)
- Koda: Advisory (you approve execution before it happens)
Human-in-the-Loop: The Core Principle
PerpDesk ensures the trader is always the final decision-maker:
What you control:
- Trade approval — every recommendation can be approved, modified, or rejected
- Risk parameters — hard limits no agent can breach
- Agent autonomy — toggle each agent's independence level
- Override capability on any recommendation
- Emergency stop available at all times
What agents handle automatically:
- Market scanning and regime detection (Vanta)
- Cross-venue signal aggregation (Luma)
- Macro context monitoring (Meridian)
- Risk enforcement and guardrails (Brix)
Agents vs. Bots vs. Copy Trading
| Dimension | Traditional Bot | Copy Trading | PerpDesk Agents |
|---|---|---|---|
| Scope | Single strategy, single exchange | Replicate another trader | Coordinated team covering entire desk |
| Awareness | Own data only | Copying without understanding | Shared context across all agents |
| Adaptability | Fixed parameters | Follows without adaptation | Dynamic, regime-aware |
| Coordination | None | N/A | Structured pipeline with consensus |
| Transparency | Often black box | You see trades, not reasoning | Every decision explained with rationale |
| Control | On/off switch | Can stop copying | Granular autonomy per agent |
| Personalization | Generic parameters | Generic copy | Configured to your style and risk |
| Human Role | Operator | Passive observer | Active decision-maker and desk lead |
The key difference: PerpDesk agents are specialists that coordinate with each other, run consensus debates, and present a validated recommendation. You remain in control as the desk lead.
Trading Terminology
Funding Rate
The periodic payment traders with long positions pay to traders with short positions (or vice versa). This keeps perpetual prices aligned with spot prices. High funding rates create cost considerations and arbitrage opportunities.
Mark Price
The price used for liquidation calculations. Derived from a combination of spot price and futures price to prevent manipulation.
Liquidation Cascade
When a large position is liquidated, the forced market order can trigger other positions, creating a waterfall of closures. These create volatility spikes and often represent tactical opportunities.
Open Interest (OI)
The total number of open positions in a perp market. Rising OI + rising price suggests strong conviction. Falling OI + rising price suggests weak hands exiting. Divergences between exchanges can signal potential moves.
Basis
The difference between perpetual price and spot price. Positive basis (perps above spot) often signals bullish sentiment and high funding. Negative basis is bearish.
Multi-Exchange Trading
PerpDesk currently supports trading on Hyperliquid and Lighter, with more exchanges coming soon. A mock exchange is also available for paper trading.
When you connect exchanges, your agents coordinate across venues:
- Luma aggregates funding rates and OI data across connected exchanges plus additional data sources
- Riven factors cross-venue data into consensus recommendations
- Brix calculates risk across all connected positions
- Koda provides execution guidance for your chosen venue
Market data is sourced from multiple providers to give agents the broadest possible view of market conditions.