Spark

Imagine every vault, wallet, and trade on Hyperliquid is a dark room. You know the data is there, but finding what matters means lighting every corner one at a time — leaderboard, vault detail, positions, trades, history. Spark illuminates the whole room at once. Light wherever you look, instantly, without flipping switches. Ask a question and the answer lights up — no hunting required.

Spark is a conversational AI that understands the full Hyperliquid vault ecosystem — every vault, every wallet, every position, every trade. Ask questions in plain English, get answers backed by real data, and drill deeper with follow-ups. One conversation replaces hours of tab-hopping.

★ Pro Spark is included with Pro.

Why use Spark instead of manual research?

Vault research traditionally means jumping between pages — leaderboard to vault detail to positions to trade history, correlating the data yourself at each step. A new question means a new round of navigation.

Spark collapses that into a single action. You ask, you get an answer, you ask the next question. The context carries forward. There are no filters to set, no tabs to switch, no SQL to write.

This matters because the best investment decisions come from asking better questions, not from reading more tables. When the cost of answering “what if” drops to near zero, you explore more freely and reach conclusions faster.

How do conversations work?

Threads keep research organized

Spark groups messages into threads. Each thread is a standalone conversation with its own context. Use separate threads for separate research topics — one for vault discovery, another for your portfolio, a third for tracking a specific wallet.

Threads are automatically named based on your first message, so you can find them later without remembering what you typed. The sidebar lists every thread. Click any to switch. The trash icon deletes threads you are done with — a two-click process so you never remove one by accident.

On desktop, the sidebar is always visible. On mobile, a hamburger button opens it as an overlay.

Responses stream in as they are generated

When you send a message, Spark responds in real time. Text streams in word by word — you never wait for the full answer before seeing the first results. While the AI works through a complex question, a “Thought” section shows its chain of reasoning so you can follow the logic.

If Spark needs to look up data, you see indicators as it works — “Querying database” while it runs a lookup, “Rendering chart” while it builds a visualization. Each step appears as it happens.

If the response is going in the wrong direction, click the Stop button to abort. The partial content stays visible.

Answers adapt to the question

Spark responds with whatever format fits the question:

Direct answers — plain-English explanations backed by real numbers
Data tables — structured results with column headers, scrollable for long results, with row counts so you know the scale
Charts — bar, line, area, horizontal bar, and pie charts rendered inline

A single question can produce all three. Ask “which vaults have the best risk-adjusted returns” and Spark might return a table of the top 10, followed by a bar chart comparing them, with a summary sentence explaining what the data means. No toggling between views.

Use voice input

Spark supports voice input through your browser. Click the microphone icon next to the input bar, speak your question, and the transcript appears in the textbox. Review it, edit if needed, and send. Useful when you are thinking faster than you can type.

Copy anything

Every message has a copy button — hover to reveal it, click to copy the full content. Code blocks inside responses have their own copy button. The “Thought” reasoning section does too. Copied content can be pasted into notes, shared with others, or used to build your own research records.

What can you ask Spark?

Vault discovery

“Which vaults have the best risk-adjusted returns over the past 90 days?”
“Show me vaults with low correlation to each other — I want to diversify.”
“Which vaults have the most consistent weekly returns with the least drawdown?”
“Find vaults that perform well during choppy markets, not just trending ones.”
“Which vaults have depositor bases that are well distributed — not dominated by one whale?”
“Which vaults created in the last 30 days are outperforming established ones?”
“Show me vaults with high TVL and good APR — what positions are they holding and how risky are they?”
“Which vaults have the most unique depositors?”

Wallet and portfolio analysis

“What does my overall portfolio look like — total equity, weighted APR, and net exposure?”
“Show me how my vault allocations have changed over the last month.”
“Which of my vault positions is dragging on returns the most?”
“What’s my net directional bias across all my vaults?”
“Compare my portfolio’s performance against the top 10 vaults on the leaderboard.”
“Which wallets have the best annualized ROI with at least $5,000 deployed?”
“Which wallets have deployed the most total capital across all vaults?”
“Which wallets have the highest win rate — the most winning vaults vs total?”

Strategy and behavior

“Which vault leaders consistently size positions well during high volatility?”
“Show me vaults that reduce leverage after drawdowns instead of doubling down.”
“Which vaults have the most predictable trading patterns — low style drift?”
“Find vaults where the leader’s actual behavior matches their stated strategy.”
“Which vaults have the fastest drawdown recovery times?”
“Which vault leaders manage the most vaults and how does their average APR compare?”
“Which vault leaders charge the highest commission — and are their vaults still performing?”

Trade and performance analysis

“What’s the average hold time for positions across all vaults?”
“Show me the largest winning and losing trades this week across every vault.”
“Which coins are vaults most bullish on right now?”
“What’s the long/short split across all vaults — are they net long or net short?”
“Which vaults are generating the most fee-efficient returns?”
“Which coins have generated the most trading volume and PnL across all vaults?”
“Show me aggregate stats — average APR, TVL, drawdown, and depositor count across all vaults.”

Risk and red flags

“Which vaults have unusually high leverage compared to their own history?”
“Show me vaults where a single depositor controls most of the capital.”
“Which vaults have gone through the deepest drawdowns and how did they recover?”
“Are there vaults that have been steadily losing equity over the past month?”
“Show me vaults with sudden changes in trading velocity or position sizing.”
“Which vaults are most reliant on a single depositor?”
“Which vaults are running the highest leverage and what coins are they betting on?”

Custom research

“Compare vault X, vault Y, and vault Z side by side on APR, drawdown, and depositor distribution.”
“Which vaults match my criteria — at least 3 months of history, under 15% max drawdown, and over 20% APR?”
“Chart the cumulative PnL of the top 5 vaults over the last 6 months.”
“What would a portfolio of 5 uncorrelated vaults look like — simulate the blended returns?”
“Which vaults look most like vault X in terms of trading style and risk profile?”
“Show me a chart comparing the top 5 vault leaders by total value managed.”
“Which high-performing vaults charge low leader commission — best deals for depositors?”

These are examples. Spark answers whatever you ask — and if the answer sparks a new question, you follow it immediately instead of starting a new search.

How does Spark handle your data?

Spark works entirely from your local data — the same vault, wallet, and position information you see in the app’s tables and detail pages. When you ask a question, Spark queries that local database directly and computes the answer. Your data stays in your browser.

To answer questions, Spark uses a read-only SQL tool against the on-device database, along with a chart tool that renders results as visualizations. Results are capped (a limited number of rows and bytes per query) so answers stay fast and readable, and large result sets include a truncation notice. Queries are read-only — Spark never modifies your data.

How do you set up Spark?

Open Spark from the top navigation bar. On first visit, you will see a prompt to configure your AI provider — this connects Spark to the language model that processes your questions. The setup takes about a minute:

Go to Settings → AI.
Enter your provider details: a label (like “OpenAI” or “Groq”), the endpoint URL, the model name, and your API key.
Click Test Connection to verify everything works.
Click Save.

Your API key is stored locally in your browser. It is never sent anywhere except to the provider you configured, and only when you ask a question. You can clear your key at any time.

Spark works with any OpenAI-compatible provider — OpenAI, Anthropic, Groq, OpenRouter, or a local model running on your machine. The defaults point to OpenAI’s API with gpt-4o-mini, but you can switch to any model you prefer.

What changes when you use Spark?

The table below shows what changes when you switch from manual research to Spark.

Without Spark	With Spark
Pick a vault from the leaderboard, open its detail page, scan positions, check trades, repeat for each vault.	“Which vaults have the best risk-adjusted returns?” — instant answer across all vaults.
Manually track your deposits across vaults to see your total exposure.	“What does my portfolio look like?” — aggregated in one response.
Wonder if a vault is unusually risky but have no easy way to compare it.	“Show me vaults with unusually high leverage compared to their own history.”
Spot a pattern across vaults but can’t articulate it into a filter.	“Which vaults reduce leverage after drawdowns vs doubling down?”
Hesitate to explore a hypothesis because it would take too long to check.	Ask anything. If it’s wrong, you find out in seconds, not hours.

Spark does not replace your judgment. It removes the friction between having a question and finding the answer. You still decide what matters. You just spend less time hunting and more time thinking.

Settings — configure your AI provider
Signals & Intelligence — automated AI analysis per vault
Vault Detail — Ask AI — chat focused on a single vault

Portfolio

Alerts