Parasol: Autonomous AI Trading Agents for Solana

Human Conviction + AI Discipline

Version 4.0 — April 2026

Abstract

Parasol is an autonomous AI trading platform for Solana memecoins. Users deploy AI agents that continuously scan thousands of tokens in real-time, filter out manipulation, score opportunities, and execute trades with disciplined risk management while they sleep, work, or live their lives.

The core thesis: humans are good at knowing what to trade; machines are good at executing without flinching. Parasol pairs human conviction with AI discipline to solve the oldest problem in trading. The best traders have the right instincts about what to buy. The worst traders have no discipline about when to sell. Parasol exists in that gap.

The platform combines multi-source data ingestion, proprietary on-chain manipulation detection, multi-strategy scoring, adaptive learning, a custom-built execution engine with direct on-chain swaps, and live execution through non-custodial MPC wallets. Everything runs at parasol.so.

1. The Solana Memecoin Market

1.1 Scale and Velocity

Memecoins are cryptocurrency tokens created around internet culture, trends, and viral moments. Solana has become the dominant blockchain for memecoin trading due to its sub-second transaction speeds and near-zero fees.

• ~10,400 new tokens launch daily on Pump.fun alone (Chainplay, 2025)
• 1.4% graduation rate — only 1 in 71 tokens survives past the initial bonding curve (Chaincatcher, 2025)
• 98% of tokens die within 3 months, with an average lifespan of just 12 days
• Only 3% of users profit more than $1,000 from trading on Pump.fun

1.2 Why Most Traders Lose

The majority of memecoin traders lose money for predictable, well-studied reasons:

• Emotion over discipline. Buying on FOMO, holding on hope, selling on panic. Prospect theory, loss aversion, and the disposition effect are features of human cognition, not character flaws. Evolution optimized us for survival, not for selling memecoins at 4.7x.
• Being exit liquidity. Research (arXiv:2507.01963) found that 82.8% of high-return tokens show evidence of artificial price inflation. If you're buying a token that's trending because of coordinated wallets, you're providing exit liquidity for insiders.
• No risk management. No position sizing, no stop losses, no portfolio-level controls. One bad trade wipes out weeks of gains.
• Speed disadvantage. The profitable window is measured in minutes. By the time a human spots a trending token, evaluates it, and executes, the optimal entry has passed.

1.3 The Opportunity

Memecoin price action is driven by attention and liquidity flow, not fundamentals. This makes it uniquely suited to quantitative analysis of trading patterns, volume dynamics, and on-chain behavior, exactly what an AI system can monitor at scale.

The discipline gap is largest in this market. Tokens move 5-50x in hours. The emotional pressure is extreme. The difference between a disciplined exit at 4x and a hopeful hold to 0x is the entire position. In slower markets, discipline matters less because the consequences of being late are smaller. In memecoins, the discipline gap is the alpha.

2. Human Conviction + AI Discipline

2.1 The Problem with Bots

Traditional trading bots eliminate discipline problems by also eliminating conviction. They replace human judgment with parameterized rules, and in a market as fast, noisy, and narrative-driven as memecoins, parameterized rules break constantly.

A bot can't tell you that a token's narrative has shifted from genuine community to paid promotion. It can't recognize that a developer's wallet pattern looks like an exit setup. It can't feel the market sentiment shift from euphoria to exhaustion.

2.2 The Parasol Approach

Parasol is not a bot. It's a collaboration between human intuition and machine execution.

What the human provides:

• Strategy selection that matches their market thesis
• Risk parameters that reflect their financial situation
• Conviction overrides when they see something the data can't see

What the AI provides:

• 24/7 execution across every market condition
• Emotionless exits when the plan says sell
• Sub-second speed from signal to on-chain execution
• Consistent application of rules across hundreds of trades
• Multi-token attention on every price tick simultaneously
• Adaptive learning that adjusts strategy weights based on performance data

2.3 The Collaboration Loop

The system gets better over time because both sides learn. The human learns what market conditions favor which strategies. The AI learns which exit rules capture the most profit, which entry filters avoid the most rugs, and how to weight competing signals. Neither side alone produces this outcome.

3. The Parasol Platform

3.1 AI Agents

Parasol users deploy AI agents, each with a distinct trading approach. Agents operate across multiple data sources and strategies:

PumpFun-focused agents target tokens on the bonding curve and recently graduated tokens. These are ultra-early-stage opportunities with the highest risk and highest potential return.

Trending-focused agents target tokens that have already gained traction on DEX aggregators. Lower risk with more established price action to analyze.

Oracle agents learn from the best trades across all other agents, synthesizing patterns from multiple strategies and market conditions.

Each agent operates independently with its own portfolio, positions, and performance history. Users can run multiple agents simultaneously, comparing strategies in real-time.

3.2 Multi-Source Data Layer

Parasol ingests data from multiple providers with automatic failover and intelligent routing:

• Token discovery across new launches, trending tokens, and graduated pairs
• Real-time pricing with sub-second updates and redundant backup feeds
• On-chain data including holder distribution, sniper detection, and developer wallet activity
• Token quality scoring using composite metrics that filter before trading
• Real OHLCV candlestick data for accurate technical analysis

If any data source goes stale or unavailable, backup sources activate within seconds. The agent always has current data to make decisions.

3.3 Five-Phase Pipeline

Every token passes through the same pipeline regardless of source:

1. Scan — Multi-source ingestion detecting tokens within seconds of launch or trend.

2. Filter — Six-layer manipulation detection. Tokens that fail any critical check are rejected before any trade is considered.

3. Score — Multiple strategies score each token in parallel. The highest-confidence strategy determines the trade setup.

4. Execute — Proprietary execution engine with direct on-chain swaps, premium network infrastructure, and multi-layer fallback paths. Optimized for >95% transaction landing rate.

5. Protect — Real-time exit evaluation on every price tick, velocity-aware trailing stops, take-profit laddering, time-based exits, daily loss limits, circuit breakers, and anti-exploitation randomization.

4. Manipulation Detection

This is the most important layer in the system. The edge comes from what you avoid, not just what you buy.

4.1 Six Categories of Risk Signals

• Holder concentration — If top wallets hold a disproportionate share of supply, insiders can crash the price at any time.
• Fresh wallet clusters — Multiple new wallets buying simultaneously indicates coordinated manipulation (Sybil attacks).
• Same-block sniping — Wallets buying in the same block as token launch indicates deployer-funded extraction. Research shows 87% of same-block snipes are profitable for the sniper.
• Extreme buy ratios — Buy transactions far exceeding sells (95%+) often indicates wash trading rather than genuine demand.
• Volume/liquidity mismatch — Trading volume vastly exceeding available liquidity is a sign of artificial activity.
• Wash trading patterns — Repetitive buy/sell cycles between related wallets designed to inflate volume metrics.

4.2 The Philosophy

The system would rather miss a legitimate opportunity than provide exit liquidity for insiders. False negatives (missed winners) are acceptable. False positives (buying into a rug) are not.

5. Trading Strategies

5.1 Core Strategies

Momentum — Identifies tokens with genuine multi-timeframe trend alignment, where volume, price action, and buying pressure all confirm upward movement.

Mean Reversion — Detects oversold conditions on established tokens where short-term dips represent buying opportunities within a healthy longer-term trend.

Breakout — Captures explosive volume surges and one-sided buying pressure that signal the beginning of a major price move.

5.2 Bonding Curve Strategies

For PumpFun tokens still on the bonding curve, traditional technical analysis doesn't apply. Parasol uses alternative signals including freshness scoring, bonding curve progress, holder quality metrics, and market cap gates.

5.3 Adaptive Learning

Parasol captures a detailed snapshot of market conditions at every trade entry and correlates features with outcomes. The system learns which characteristics predict winners vs. losers, which strategy performs best in current conditions, and which token profiles correlate with positive outcomes.

Learning operates at three levels: global (all trades), regime-specific (bull/neutral/bear markets), and regime+strategy-specific. Weight adjustments require minimum sample sizes to prevent over-fitting.

6. Execution Engine

6.1 Proprietary Direct Swaps

Parasol does not rely solely on third-party DEX aggregators. We built a custom execution engine that goes directly to the source for PumpFun ecosystem tokens, reading on-chain state, calculating outputs locally, and submitting optimized transactions through premium infrastructure.

For tokens on other DEXes, we use the latest managed execution APIs with intelligent retry and fee escalation. Multiple execution paths run simultaneously, with the first successful confirmation winning. This dual-path approach achieves landing rates competitive with the best trading terminals on the market.

6.2 Transaction Reliability

Every aspect of the execution stack is optimized for reliability:

• Premium network infrastructure with stake-weighted quality of service
• Multi-path submission with simultaneous execution through independent channels
• Smart routing that auto-detects token state and picks the optimal execution path
• Intelligent retry with escalating priority fees and fresh transaction rebuilds
• Circuit breakers that automatically pause trading if failure rates spike
• MEV protection through atomic bundle submission on all sells

6.3 Non-Custodial Wallets

Parasol executes live trades through MPC (Multi-Party Computation) wallets powered by Turnkey. Users' private keys are never stored or accessible by Parasol. The wallet is created during onboarding and controlled entirely by the user.

6.4 Paper vs. Live

Every agent can run in paper mode (simulated) or live mode (real funds). Paper mode is the default. Users observe agent performance before committing real capital. Switching to live mode requires wallet connection and fund deposit.

7. Risk Management

7.1 Position-Level Controls

• Calculated position sizing based on portfolio size, confidence, and risk profile
• ATR-based trailing stops that adapt to each token's volatility
• Take-profit laddering that captures gains at multiple levels
• Moon bag engineering that keeps small positions running after taking profit

7.2 Exit Intelligence

Parasol's exit system operates on every price tick, not on a fixed schedule:

• Velocity-aware trailing stops that tighten as momentum decays and widen when momentum is strong
• Momentum-based partial exits that lock in profit when trends shift while keeping exposure to continued upside
• Anti-exploitation randomization on all exit parameters to prevent adversarial bots from predicting the agent's behavior
• Acceleration detection that identifies inflection points where pumps turn into dumps

7.3 Portfolio-Level Controls

• Diversification across many concurrent small positions
• Daily loss limits that halt all trading if cumulative losses reach a threshold
• Time-based exits that close stale positions
• Cross-agent cooldowns preventing re-entry into recently closed tokens
• Closed position memory ensuring agents never re-enter dead tokens
• Token banning allowing users to permanently exclude any token

7.4 The Asymmetric Bet Structure

• Typical loss: Capped by trailing stops at a defined percentage
• Typical win: Take-profit laddering captures gains at multiple levels
• Rare big win: Trailing stops on remaining portions ride exceptional moves

With many small positions and disciplined exits, even a modest win rate is profitable when average wins exceed average losses.

8. Security

8.1 Authentication

Every API endpoint that modifies state requires authentication via signed session tokens (HMAC-SHA256, 7-day TTL). Token verification uses constant-time comparison to prevent timing attacks. Legacy authentication fallbacks have been removed.

8.2 Wallet Security

• Non-custodial MPC wallets via Turnkey, institutional-grade key management
• Private keys never stored or accessible by Parasol
• All wallet operations require authenticated session tokens
• Rate limiting on sensitive operations

8.3 Anti-Exploitation

All trading parameters include proprietary randomization with entropy seeding. Stop-loss levels, take-profit levels, trailing stop distances, and entry timing all include intelligent noise that makes the agent's behavior unpredictable to external observers. The specifics are confidential.

8.4 Infrastructure Protection

• Circuit breakers that pause trading during anomalous conditions
• Automated safety systems monitoring transaction health in real time
• Cron endpoint protection via signed secrets
• Pre-commit hooks enforcing type safety on every change

9. Parasol Tools

Parasol provides three free trading calculators at parasol.so/tools:

Position Sizing Calculator — Calculate optimal position sizes based on account balance, risk tolerance, entry price, and stop loss. Includes live asset prices.

Funding Rate Calculator — Calculate the cost of holding perpetual futures positions over time. Integrates live funding rates from Hyperliquid.

Liquidation Price Calculator — Calculate liquidation prices for leveraged positions across isolated and cross margin modes.

10. Gamification and Community

10.1 Points System

Users earn points for executing trades (paper and live), daily logins and streaks, referring new users, and completing quests. Auto-reconciliation ensures point accuracy across all users.

10.2 Referrals

Users create personal referral codes. Referred users receive a fee rebate, and referrers earn a percentage of trading fees generated by their referrals, with multi-level attribution.

10.3 Leaderboard

A public leaderboard ranks users by points across 24-hour, 7-day, 30-day, and all-time periods.

11. Architecture

11.1 Technology Stack

• Framework: Next.js 15, React 19, TypeScript
• Styling: Tailwind CSS v4
• State: Zustand (client), Supabase PostgreSQL (persistence)
• Wallets: Turnkey MPC (non-custodial)
• Execution: Proprietary direct swap engine + Jupiter fallback
• Data: Multi-source with automatic failover and caching
• Deployment: Vercel, auto-deploys from main branch
• Testing: Vitest, pre-commit type checking via Husky
• Domain: parasol.so

11.2 Execution Architecture

The trading engine operates a multi-layer execution stack. For PumpFun ecosystem tokens, transactions are built directly against on-chain programs and submitted through premium infrastructure. For tokens on other DEXes, managed execution APIs handle routing and retry. Multiple submission paths run simultaneously to maximize landing rates.

All sells include MEV protection through atomic bundle submission. Priority fees are calculated dynamically based on current network conditions and the specific accounts each transaction touches.

12. Roadmap

Shipped

• Multi-source data layer with automatic failover
• Six-layer manipulation detection
• Multiple entry strategies including bonding curve, momentum, breakout, mean reversion
• Adaptive learning engine with stability gates
• Paper trading with full dashboard
• Proprietary direct swap engine (PumpFun bonding curve + PumpSwap AMM)
• Premium staked network infrastructure
• Dual-path transaction submission (simultaneous multi-channel execution)
• Reactive exit system with per-tick evaluation
• Velocity-aware trailing stops with momentum tracking
• Anti-exploitation randomization on all trading parameters
• Circuit breaker wallet protection
• 100% authentication coverage on all mutation endpoints
• MEV protection on all sells
• Named AI agents with distinct strategies
• Agent arena for running multiple agents simultaneously
• Gamification (points, referrals, leaderboard)
• Free trading tools (position sizing, funding rate, liquidation price)
• Blog with regular product updates

In Progress

• Momentum decay partial exits (sell 50% when trend shifts, keep the rest running)
• Deterministic take-profit execution
• Moon bag engineering for capturing extended runners
• Intelligence surfaces (sniper badges, token scores, holder trends on position cards)
• Partial sell UI (25/50/75/100% buttons)
• Telegram notifications for trade alerts

Planned

• Subscription tiers (free paper trading, paid live features)
• Dedicated always-on server infrastructure for real-time monitoring
• Multi-user architecture for scaling beyond early adopters
• Advanced portfolio analytics and performance reporting
• Mobile-optimized experience
• Token deployment capabilities

13. Frequently Asked Questions

What is Parasol?

A platform where you deploy AI trading agents that autonomously trade Solana memecoins. Each agent has its own strategy and portfolio. You can run them in paper mode (simulated) or live mode (real funds).

How is Parasol different from a trading bot?

Bots execute fixed rules. Parasol agents learn, adapt, and make intelligent decisions based on real-time market data. The system pairs your market intuition with machine-grade execution discipline.

Is Parasol free to use?

Paper trading is free. Live trading involves a small platform fee on trades. Subscription tiers for premium features are coming soon.

Are my funds safe?

Parasol uses non-custodial MPC wallets powered by Turnkey. Your private keys are never stored or accessible by Parasol. You maintain full control of your funds at all times.

How does Parasol detect rug pulls?

Six categories of on-chain analysis: holder concentration, wallet age patterns, sniping activity, volume/liquidity ratios, buy/sell pressure dynamics, and wash trading detection. Tokens that fail any critical check are rejected before any trade is considered.

How reliable are the trades?

We built a proprietary execution engine that goes directly to the source rather than relying solely on third-party aggregators. Combined with premium network infrastructure and multi-path submission, our target is >95% first-attempt landing rate on sells, competitive with the best trading terminals on Solana.

Can I run multiple agents at once?

Yes. Each agent operates independently with its own portfolio. You can diversify across different strategies and data sources simultaneously.

14. References

1. Chainplay. "Every 24 Hours on Pump.fun, 10,417 Tokens Are Launched while 9,912 Become Defunct." 2025.
2. Chaincatcher. "The real data of pump.fun revealed: a graduation rate of 1.4%, and only 3% of users profit over 1000 dollars." 2025.
3. arXiv:2507.01963. "A Midsummer Meme's Dream: Investigating Market Manipulations in the Meme Coin Ecosystem." 2025.
4. arXiv:2412.20138. "TradingAgents: Multi-Agents LLM Financial Trading Framework." Tauric Research, 2024.
5. Messari. "Systematic Memecoin Investing." 2025.
6. Galaxy Digital. "The State of Memecoins: Culture, Trading, and Infrastructure." 2025.
7. Pine Analytics. "Exit the Liquidity Machine: Unveiling the Internal Sniping Arbitrage of Pumpfun Token Issuance." 2025.
8. Stratium. "Solana Meme Coin Trading Strategies: How Top Traders Profit." 2025.

Parasol is experimental software. Memecoin trading is extremely high-risk. 98% of memecoins fail. Past performance does not predict future results. Never trade with money you cannot afford to lose.