Frequently Asked Questions

Lucy is an autonomous AI trader powered by frontier AI models from Anthropic (Claude), Google (Gemini), and DeepSeek. She operates 24/7, analyzing cryptocurrency markets and executing trades independently based on your chosen strategy. Unlike simple trading bots, Lucy uses genuine artificial intelligence to interpret market conditions, learn from outcomes through vector memory, and make contextual decisions that adapt over time.

Traditional bots follow rigid if-then rules. Lucy uses advanced AI reasoning with "extended thinking" capability - she literally thinks through each decision before acting. She interprets strategy frameworks intelligently, adapts to changing market conditions, and learns from her trades through RAG (Retrieval-Augmented Generation) vector memory. Each trading decision goes through Claude or Gemini's reasoning engine, considering multiple factors simultaneously.

No. Lucy is a full autonomous execution system. She connects directly to your exchange account, analyzes markets, decides when to trade, and places orders herself. No manual intervention required - no copying signals, no following alerts. Lucy makes independent decisions based on your strategy configuration.

Lucy runs continuously without human intervention. She monitors markets 24/7, maintains her own trading logic, manages positions, and adapts to market conditions. You can chat with her through the dashboard to adjust strategies or ask questions, but she doesn't need you to approve each trade.

Lucy has access to 20+ technical indicators including RSI, MACD, ADX, Bollinger Bands, Keltner Channels, TD Sequential, EMAs, ATR, pivot points, and Fibonacci levels - across 6 timeframes (1m, 5m, 15m, 1h, 4h, daily). She requests only relevant indicators for each decision, keeping analysis efficient. She also monitors order book depth, funding rates, open interest, and CVD (cumulative volume delta).

A pulse is Lucy's heartbeat - a scheduled moment when she wakes up to check market conditions, review positions, analyze indicators, and decide whether to act. Lucy controls her own pulse timing (30 seconds to 30 minutes) based on the situation: faster pulses (30s-2min) when price approaches her levels or during high volatility, slower pulses (10-30min) when flat and waiting for setups. Between pulses, she monitors real-time events like order fills via WebSocket, responding immediately when trades execute. You pay per pulse, so Lucy optimizes timing to be cost-efficient while staying responsive.

Yes. Lucy maintains persistent memory across thousands of pulses and learns from every outcome through vector memory. When trades close, episode memories are automatically created with market conditions, indicators, and results. At each pulse, Lucy retrieves similar past experiences using semantic search to inform her decisions. Even after restarts, her memories persist in the database.

Fill events trigger immediate response - under 100ms from WebSocket event to Lucy analyzing and potentially placing counter-orders. This is critical for grid trading where fills need immediate replenishment. Between fills, Lucy controls her own pulse timing based on market activity - as fast as 30 seconds when action is imminent, or up to 30 minutes when patiently waiting for setups to develop.

Extended thinking is the AI's ability to reason through complex problems before responding. Claude, Gemini, and DeepSeek Reasoner models all support this capability. For every trading decision, Lucy thinks through the setup, weighs multiple factors, considers risks, and forms a conclusion - all before executing. You can see her thought process in the dashboard Activity tab.

Traditional bots repeat the same mistakes forever. Lucy learns. Her vector memory automatically captures every trade outcome with full market context. At each pulse, she retrieves similar past experiences using semantic search to inform her decisions. Memories that lead to wins get promoted; memories from losses stay as cautionary examples. After 100 trades, she has 100 relevant experiences to draw from.

Lucy uses RAG (Retrieval-Augmented Generation) with vector embeddings. When a trade closes, an episode memory is automatically created containing: market conditions (RSI, trend, volatility), entry/exit prices, P&L, and outcome. These are converted to 384-dimensional vectors using a local AI model. At each pulse, Lucy's current market state is compared against all memories using semantic similarity search.

Instead of keyword matching, Lucy finds memories based on meaning. "RSI oversold in ranging market" will match past trades with similar conditions even if described differently. The embedding model understands context - a memory about "buying the dip during consolidation" is relevant when current RSI is low and ADX shows no trend, even without exact word matches.

Lucy uses a three-tier system inspired by FinMem research: Working (15 memories, new observations), Proven (10 memories, validated through wins), and Core (5 memories, highest-confidence insights). Memories start in Working, get promoted to Proven after 3+ wins with >70% confidence, and graduate to Core after 7+ wins with >90% confidence. Unreliable memories decay and are evicted.

Yes - episode memories are created automatically the moment a position closes. No manual intervention needed. Lucy captures entry/exit prices, P&L, hold time, RSI, trend direction, volatility regime, CVD (buy/sell pressure), and trading session (Asia/London/NY). This rich context helps her find truly similar situations later.

Yes. Lucy has three memory tools: remember() to store observations, recall() to retrieve memories by search or filter, and reinforce() to update outcomes after trades. She can explicitly note patterns like "breakouts failing in low volume" and later reinforce whether that insight proved correct.

Yes. The dashboard Memory tab shows Lucy's vector memories: each episode with market conditions, confidence scores, win/loss records, and layer (working/proven/core). You can watch memories get promoted as they prove reliable. It's transparent learning you can actually observe.

Yes. All memories are stored in PostgreSQL with pgvector extension. Restart her, change models, come back a week later - her memories remain intact. She picks up exactly where she left off with all accumulated experience. This is persistent intelligence, not session-based memory that evaporates.

Backtesting looks at historical data that may no longer be relevant. Lucy learns from live trades with real fills, actual slippage, and current market conditions. Her memories evolve continuously - new experiences are added, unreliable ones decay. She adapts to what's working NOW, not what worked in data from years ago.

Currently Bybit (USDT perpetual futures) and Hyperliquid (USDC perpetual futures). Both are fully integrated with real-time WebSocket connections for instant order fills, position updates, and balance tracking.

Perpetual futures contracts only. Currently single trading pair per instance (e.g., BTCUSDT, ETHUSDT, SOLUSDT). One Lucy instance equals one trading pair. This focused approach allows Lucy to develop deep expertise on specific markets.

No. Lucy only has trading permissions. She cannot withdraw, transfer, or access funds outside of placing and canceling orders. Your API keys should be configured with trading-only permissions for maximum security.

Leverage must be set manually on your exchange (Bybit web/mobile app or Hyperliquid interface) before starting Lucy. She reads the leverage setting from the exchange but cannot change it. This is a deliberate safety feature.

Yes. Each instance requires a separate Bybit sub-account or Hyperliquid wallet. One Lucy = one exchange account = one trading pair. You can run as many instances as you want, each with different strategies.

Hyperliquid uses "Agent Wallets" instead of traditional API keys. An Agent is a secondary wallet that can sign trades on your behalf but cannot withdraw your funds - keeping your main wallet secure.

1. Go to the API page
   Visit app.hyperliquid.xyz/API (or click More → API in the navigation)
2. Create your agent
   Enter a name for your agent (e.g., "Lucy") and click "Generate Wallet Address". This creates a new agent wallet locally in your browser.
3. Copy the Agent Private Key
   Copy and save the Private Key immediately! This is shown only once. You'll need this for the "Private Key" field.
4. Authorize the agent
   Set the validity period to MAX (up to 180 days). Click "Authorize" and sign the transaction with your main wallet.
5. Enter credentials in Lucy
   • Wallet Address → Your main wallet address (the one connected to Hyperliquid that holds your funds)
   • Private Key → The agent's private key you copied in step 3 (NOT your main wallet's private key!)

Bybit uses traditional API keys with permissions you control.

1. Log in to Bybit
   Go to bybit.com and log in to your account.
2. Open API Management
   Click your profile icon → "API" (or go directly to bybit.com/app/user/api-management)
3. Create a new key
   Click "Create New Key" → Choose "System-generated API Keys"
4. Set permissions
   • Enable "Contract" (required for futures trading)
   • Enable "Read" (required for checking balance and positions)
   • Do NOT enable "Withdraw" (Lucy doesn't need this - keep your funds safe!)
5. Set IP restriction
   Choose "No restriction" for easiest setup. (Advanced users can add specific server IPs)
6. Complete verification
   Enter your Google Authenticator 2FA code to confirm.
7. Save your credentials IMMEDIATELY
   • API Key → paste in the "API Key" field
   • API Secret → paste in the "API Secret" field

Yes. Strategies are written in Markdown format with sections for entry/exit rules, position sizing, risk limits, and key indicators. Lucy interprets the strategy intelligently rather than following rigid rules. You describe your trading philosophy, and she applies it contextually.

Lucy excels at grid trading, mean reversion, and momentum strategies. Strategies that benefit from continuous monitoring, quick reaction to fills, and adaptive position sizing work particularly well. She's especially effective at managing multiple grid levels and responding to partial fills.

Strategies should describe your trading philosophy, key indicators to watch, entry/exit criteria, position sizing rules, and risk limits. Lucy works best with clear guidelines but room for interpretation. Overly rigid rules limit her ability to adapt; too vague and she may not capture your intent.

Yes - this is one of Lucy's most powerful features. Her vector memory automatically captures every trade with full market context. At each pulse, she retrieves similar past experiences using semantic search. Memories that prove reliable get promoted; unreliable ones decay. Over time, she develops genuine expertise in your specific market. See the "Vector Memory" section for full details.

Yes. Lucy is not just an executor - she's a trading partner. You can chat with her to analyze market conditions, discuss strategy ideas, backtest concepts, and refine your approach. She can explain her reasoning, reference similar past trades from her memory, and help you iterate on your strategy in real-time.

Yes. Strategies can be shared with other users on the platform. You can also make your Lucy instance public, allowing others to watch her trades and activity in real-time (view-only - they cannot control your instance or see your credentials). This is great for transparency, education, or building a following.

API keys are encrypted with AES-256-GCM before storage. The encryption key never leaves the server. Keys are decrypted only when Lucy's isolated worker process needs them - never exposed to the web interface or stored in plaintext.

Change them immediately in the dashboard and on your exchange. Lucy cannot withdraw funds, so worst case is unauthorized trades. New keys take effect on instance restart. We recommend using sub-accounts with trading-only permissions and IP restrictions.

Only you. Each instance is completely isolated. Your positions, orders, balance, chat history, and strategy are private to your account. We don't share, sell, or analyze user trading data.

No. Anthropic's API has a zero-retention policy for business usage. Your trading data, conversations with Lucy, and strategy details are not used for model training by either us or Anthropic.

Pay-as-you-go with Bitcoin via Lightning Network. You deposit satoshis (sats) and we deduct based on actual AI token usage. No subscriptions, no minimums, no commitments. You only pay for compute time Lucy actually uses.

Costs depend on trading activity and model choice. Each 5-minute pulse costs: DeepSeek ~10 sats, Gemini Flash ~25 sats, Haiku/Gemini Pro ~45 sats, Sonnet ~60 sats, Opus ~100 sats. With 12 pulses/hour, DeepSeek costs ~120 sats/hour. Prompt caching reduces costs significantly. At $100k BTC, 1,000 sats = $1.

You can choose between Claude (Anthropic), Gemini (Google), and DeepSeek models. Claude Opus 4.5 offers the deepest reasoning for complex strategies. Claude Sonnet balances capability and cost. Gemini 3 Flash offers excellent quality at lower cost. DeepSeek provides the best value for money - great for budget-conscious traders. You can switch models anytime from the dashboard.

Deposit sats via Lightning Network directly from the dashboard. Scan the QR code with any Lightning wallet (Phoenix, Muun, Wallet of Satoshi, etc.). Funds arrive instantly - no confirmations needed.

Lucy pauses automatically. Your positions remain open on the exchange, but Lucy stops analyzing and trading. Add more sats and restart to resume. No data is lost - she picks up exactly where she left off.

Yes. Lucy is a trader, not a guarantee. Markets are unpredictable, and even the best strategies have losing periods. Built-in safeguards exist (daily loss limits, consecutive loss pauses) but cannot eliminate risk. Start with small positions and funds you can afford to lose.

Multiple layers: Strategy-defined risk limits (daily loss limits, position sizing rules). Lucy follows her strategy's risk guidelines using AI judgment. No withdrawal access - trading permissions only. You can stop Lucy instantly from the dashboard. Her memory system helps her learn from losses and avoid repeating mistakes.

Single trading pair per instance. Perpetual futures only (no spot trading). Leverage must be set manually on exchange. Requires active strategy configuration. Benefits from occasional monitoring (not completely "set and forget"). Cannot predict black swan events.

The server is designed for high availability with automatic worker restart. If the entire platform goes down, your positions remain on the exchange with whatever stop-losses you've set. Lucy will resume trading when service returns with full state recovery from the database.

1. Create an account at lucy.guru. 2. Add your exchange API keys (trading permissions only). 3. Choose or create a trading strategy. 4. Set your leverage on the exchange. 5. Deposit sats via Lightning. 6. Start your Lucy instance. She'll begin analyzing immediately.

No. Strategies are written in plain Markdown/English. The dashboard handles everything else. If you can describe your trading rules in writing, you can use Lucy. No programming, no scripting, no technical setup required.

Yes. Use Bybit's testnet with test API keys. Lucy doesn't distinguish between testnet and mainnet - she'll trade whatever account you connect. Alternatively, start with minimal position sizes on mainnet to limit risk while learning.

Through the dashboard Chat tab. You can ask Lucy questions, give feedback on trades, request strategy adjustments, or just chat. She responds in real-time and incorporates your feedback into her decision-making.

Lucy supports multiple frontier AI models: Claude Opus 4.5, Sonnet, and Haiku from Anthropic; Gemini 3 Flash and Pro from Google; and DeepSeek Chat and Reasoner. Most models support extended thinking for deep analysis. Claude Opus offers the most sophisticated reasoning; DeepSeek provides the best value for money. You choose the model when creating an instance.

Lucy uses a distributed backend with Redis + BullMQ for job queuing. The server manages WebSocket connections to exchanges, handles authentication, and routes events. Each Lucy instance runs pulses via stateless workers that process jobs from the queue. PostgreSQL stores all persistent state, while Redis handles real-time coordination and caching.

Each Lucy instance has a permanent system prompt (lucy-core.xml) that defines her personality, trading tools, risk management rules, and behavioral guidelines. Your strategy is injected alongside this core prompt. The system prompt is carefully engineered to make Lucy a responsible, thoughtful trader who explains her reasoning and respects risk limits.

Lucy has 23 specialized trading tools: market data (trend, volatility, volume, levels - 20+ indicators across 6 timeframes), order management (place, modify, cancel), position tracking, balance queries, orderbook analysis, volume delta (CVD), memory operations (remember, recall, reinforce), messaging, and pulse control. Each tool is type-safe and validated before execution.

Events are processed through a FIFO queue with execution locks preventing race conditions. Fill events are grouped in 200ms windows to batch multi-fill orders. This ensures Lucy never places conflicting orders from parallel processing.

Lucy maintains persistent WebSocket connections to exchanges for real-time data. Public streams provide price/candle updates; private streams deliver fills, position changes, and balance updates. Connections auto-reconnect with exponential backoff. Each exchange has a dedicated client with normalized event formatting.

PostgreSQL with pgvector extension stores all persistent data: chat history, pulse outputs, trade records, vector memories (384-dimensional embeddings), user accounts, and encrypted API credentials. The schema uses Drizzle ORM with full TypeScript types. Each instance's data is isolated by instance_id.

Lucy runs 24/7 on dedicated infrastructure. The server maintains WebSocket connections and automatically restarts workers if needed. Server restarts are invisible - Lucy's pulse counter continues from where it left off with full context preservation from PostgreSQL.

The dashboard is built with Svelte 5 (latest runes syntax), TypeScript, Tailwind CSS v4, and shadcn-svelte components. Real-time updates flow through WebSocket connections to the Guardian. The app is a PWA (Progressive Web App) installable on mobile devices.

Not currently. Lucy requires specific infrastructure (Guardian server, PostgreSQL, WebSocket connections, encrypted credential management) that's handled by our platform. Self-hosting may be available in the future for enterprise customers.

Join our official Telegram group at https://t.me/lucy_guru - this is the best place to ask questions, report issues, share feedback, and connect with other Lucy traders. For announcements and updates, follow us on Nostr at https://primal.net/lucy_guru.

Yes! Our Telegram group (https://t.me/lucy_guru) is an active community where traders share strategies, discuss market conditions, help each other with setup questions, and share their Lucy experiences. Everyone from beginners to experienced traders is welcome.

Nostr is a decentralized social protocol popular in the Bitcoin community. We use it because Lucy is Bitcoin-native (Lightning payments) and we believe in decentralized, censorship-resistant communication. Follow us at https://primal.net/lucy_guru for updates, announcements, and insights.

Post in our Telegram group (https://t.me/lucy_guru) with details about the issue or your feature idea. Include screenshots if relevant. We actively monitor the group and respond to all feedback. For urgent issues, tag the admin directly in the group.