The AI Chat lives in the Workbench tab of the Thrive platform. It looks like a chat window. You type a message, hit send, and the AI responds. The responses stream in real-time so you see results generating, not just a loading spinner followed by a wall of text.
General-purpose AI chatbots can talk about trading, but they cannot touch your data. They cannot query your actual trade history, run statistical tests against real market data, or produce visualizations from live databases. They hallucinate numbers. They give generic advice based on training data that might be years old.
The Thrive AI Chat is different because it has direct access to your data and purpose-built analysis tools. When you ask "What is my win rate on short trades during volatile markets?" the AI does not guess. It generates a SQL query against your actual trades table, joins it with market regime data, filters for short trades in volatile periods, calculates the win rate, and returns the exact number with context.
The responses are grounded in your real data, executed against real databases, using validated statistical methods. This is not AI-generated content. It is AI-driven analysis.
Messages support up to 3 image attachments for visual analysis (more on that later). You can edit and resubmit previous messages to refine your questions. Queries can be re-run if you want updated results with fresh data. And the system maintains conversation context, so follow-up questions build on previous answers without needing to re-explain the setup.
The AI Chat supports four modes, each optimized for a different type of analysis.
The default mode intelligently routes your query to the right approach based on complexity. Simple questions get fast answers. Complex questions trigger multi-step analysis. You do not need to choose—the AI figures out what your question requires.
Fast, focused answers for straightforward questions. "What is BTC's current funding rate?" or "How many trades did I make last week?" These are quick data lookups that return in seconds. Low cost for simple needs.
Comprehensive multi-step analysis for complex questions. "Analyze my trading performance across all market regimes and identify where my edge is strongest" triggers multiple queries, statistical tests, and visualizations woven into a detailed report. The AI breaks the problem into steps, executes each one, and synthesizes the results into actionable insight.
This mode is expensive because it consumes significant computational resources. But for questions that would take you hours to answer manually—or that you could not answer at all without coding—200 credits is a bargain.
Plan-first execution for multi-step workflows. When you ask a complex question, Build Mode first generates an execution plan showing the steps it will take, then waits for your approval before running them. This gives you visibility into the analytical process and the ability to modify the approach before execution.
Build Mode is useful when you want to understand the methodology, not just the result. If the AI proposes a analysis approach you disagree with, you can redirect it before it spends credits on the wrong analysis.
The most common thing the AI does under the hood is generate and execute SQL queries. You ask a question in English, the AI writes the appropriate SQL, runs it against the database, and returns the results in a formatted table with interpretation.
You type: "Show me my top 10 most profitable trades in the last 90 days"
The AI generates:
SELECT
symbol,
direction,
entry_date,
entry_price,
exit_price,
ROUND(pnl_percent, 2) AS pnl_pct,
ROUND(pnl_usd, 2) AS pnl_usd,
ROUND(position_size, 2) AS size
FROM trades
WHERE entry_date >= NOW() - INTERVAL '90 days'
AND pnl_usd IS NOT NULL
ORDER BY pnl_usd DESC
LIMIT 10
And returns:
| symbol |
direction |
entry_date |
entry_price |
exit_price |
pnl_pct |
pnl_usd |
size |
| BTC/USDT |
LONG |
2026-01-19 |
96,300 |
99,800 |
3.63 |
1,815 |
50,000 |
| ETH/USDT |
LONG |
2026-01-12 |
3,180 |
3,410 |
7.23 |
1,446 |
20,000 |
| SOL/USDT |
SHORT |
2026-01-28 |
148.50 |
132.20 |
10.97 |
1,097 |
10,000 |
| BTC/USDT |
LONG |
2026-01-28 |
102,450 |
104,890 |
2.38 |
952 |
40,000 |
| LINK/USDT |
LONG |
2026-01-08 |
18.40 |
21.20 |
15.22 |
912 |
6,000 |
Along with an interpretation: "Your most profitable trades cluster in LONG positions on BTC and ETH during the January recovery. Your SOL short stands out as the only profitable short trade in the top 5, suggesting your short-side edge is more selective. The average position size on your top winners is $25,200, which is 26% larger than your overall average, indicating you size up appropriately on higher-conviction setups."
You got the data, the visualization, and the tactical insight from a single sentence. That same analysis in a traditional workflow would require: opening a spreadsheet, exporting trade data, sorting by P&L, creating a table, and manually analyzing the patterns. In the Workbench, it takes 10 seconds.
Because the AI maintains context, you can ask follow-up questions:
"What market regime were each of these trades in?"
The AI joins the existing results with regime data and adds a column showing whether each trade occurred during trending, ranging, or volatile conditions. No need to re-explain which trades you are referring to.
"Show me a chart of my cumulative P&L from these trades"
The AI generates a visualization cell with an area chart showing the running total of profits from your top 10 trades.
This conversational flow mirrors how you would work with a human analyst. Ask a question. Get an answer. Ask a follow-up. Drill deeper. Each response builds on the last.
The AI Chat has access to all 15 quantitative analysis tools in the Workbench. You do not need to know the names of these tools or how to configure them. You describe what you want to understand, and the AI selects and runs the appropriate tool.
-
You ask: "Is there a correlation between funding rate and my trade outcomes?"
-
The AI runs: Correlation Analysis tool (5 credits), computing Pearson and Spearman correlation between your trade entry funding rates and trade P&L percentages.
-
You get: "Pearson correlation: -0.34 (moderate negative). Spearman correlation: -0.41 (moderate negative). Interpretation: your trades perform better when funding rates are lower (more negative). This suggests you have a contrarian edge—you profit when the market is positioned bearishly. Consider weighting toward long trades when funding is below -0.01."
-
You ask: "Does whale buying actually predict BTC price movement, or do they just correlate?"
-
The AI runs: Granger Causality Test (10 credits), testing whether whale accumulation Granger-causes BTC returns at 1-5 day lags.
-
You get: A table showing statistical significance at each lag, with interpretation explaining that whale accumulation Granger-causes BTC returns at 2-4 day lags (p < 0.01), but not at 1 day or 5 days. The AI recommends using whale accumulation as a signal with a 2-3 day lead time.
-
You ask: "What market regime are we in right now for BTC?"
-
The AI runs: Market Regime Detection tool (5 credits).
-
You get: "Current BTC regime: TRENDING UP. Classification confidence: 87%. The current trending regime began approximately 12 days ago. Average historical duration of uptrending regimes: 34 days. Based on duration alone, the current trend is in its early-to-mid phase. Strategies that perform best in trending regimes include momentum-following entries and trend breakout positions."
None of these interactions required knowing what a Granger test is, how to calculate Spearman correlation, or what parameters regime detection models need. The AI translates your plain-language question into the appropriate statistical method and returns an interpreted result.
One of the most powerful use cases for the no-code AI is analyzing your own trading performance. Most traders know their win rate and maybe their average P&L. The AI digs into seven dimensions of performance attribution that reveal where your money actually comes from.
"Find my alpha"
The AI runs a comprehensive trade analysis covering your profitability profile, regime breakdown, time-based performance, and signal correlation. It identifies your specific edge: perhaps you consistently profit on long trades during trending markets with a 73% win rate, but your short trades in ranging markets lose money at a 62% loss rate. The recommendation is clear: lean into what works, cut what does not.
"What predicts my P&L?"
The AI runs multi-factor regression (10 credits) against your trades, testing which available data features correlate with your trade outcomes. The output shows feature importance: maybe your trade P&L is most predicted by market regime (28% importance), time of day (22% importance), and position size relative to average (18% importance). This tells you that when you trade matters more than what you trade.
"Is my edge degrading?"
The AI runs Alpha Leak Detection (5 credits) and rolling expectancy analysis. The output shows your trailing 30-trade win rate and expectancy over time. If the line is declining, your edge is eroding and you need to adapt. If it is stable, your strategy is holding up. If it is improving, you are in a favorable regime for your approach.
"How should I size my next trade?"
The AI analyzes your recent volatility exposure, current drawdown position, and win rate to recommend a position size. If you are in a drawdown, it suggests reducing size. If you are on a winning streak with stable metrics, it suggests maintaining or slightly increasing. The recommendation is based on your actual performance data, not generic rules.
The AI Chat does not just produce text and tables. It generates full visualizations when appropriate.
When you ask questions that naturally produce visual output, the AI generates charts automatically:
"Show me my equity curve for the last 6 months" → Area chart of cumulative P&L over time
"Plot my win rate by day of week" → Bar chart showing win rate for Monday through Sunday
"Create a heatmap of my performance by hour and day" → Heatmap with hours on one axis and days on the other, color-coded by average P&L
"Chart BTC funding rate versus price for the last 30 days" → Dual-axis line chart with price and funding overlaid
Each visualization appears inline in the chat and can be pinned to a dashboard for persistent monitoring. The chart types are automatically selected based on the data structure—the AI knows that time series data works best as line charts, categorical comparisons work best as bar charts, and two-dimensional relationships work best as heatmaps.
For single-metric questions, the AI generates counter widgets:
"What's my win rate this month?" → Counter showing "67.4%" with comparison to last month
"How much have I made today?" → Counter showing "+$482.50" with color coding (green for positive)
These counters are immediately pinnable to dashboards for at-a-glance monitoring.
The AI Chat accepts up to 3 images per message. This means you can screenshot a chart, a trade setup, or a market condition and get instant analysis without describing what you are looking at.
-
Chart Pattern Analysis: Screenshot a candlestick chart from any platform and ask "What pattern do you see here and what does it suggest?" The AI identifies patterns, support/resistance levels, and potential setups.
-
Trade Review: Screenshot your exchange's trade history and ask "What patterns do you notice in my recent trades?" The AI analyzes entry timing, sizing patterns, and outcome clusters visible in the screenshot.
-
Indicator Interpretation: Screenshot a chart with unfamiliar indicators and ask "What are these indicators telling me?" The AI identifies the indicators and interprets their current readings.
-
Market Structure: Screenshot a multi-timeframe view and ask "What is the market structure across these timeframes?" The AI provides a multi-timeframe analysis based on what it observes.
This feature is particularly valuable for traders who are still developing their analytical skills. Instead of wondering what a chart pattern means, you get an immediate expert interpretation grounded in the actual visual data.
Build Mode is designed for complex analytical questions that require multiple steps to answer properly. Instead of the AI running everything in one shot, it first generates a plan, shows it to you, and executes step-by-step after your approval.
You ask: "I want a complete analysis of whether funding rate mean reversion is a viable strategy for BTC over the past year."
The AI generates an execution plan:
- Pull BTC funding rate history for the past 12 months
- Calculate forward returns at 4h, 12h, 24h, and 48h horizons after extreme funding events
- Run correlation analysis between funding extremes and forward returns
- Test Granger causality to confirm prediction vs coincidence
- Segment results by market regime to identify when the signal works best
- Measure signal decay across time horizons
- Build a preliminary backtest with entry on extreme negative funding and exit on normalization
- Run Monte Carlo simulation on the backtest results
- Generate a comprehensive report with visualizations
You review the plan. Maybe you want to add "also test extreme positive funding for short entries." You modify the plan and approve. The AI then executes each step sequentially, showing intermediate results as it goes.
Build Mode is the closest thing to having a quant analyst on staff. You define the question, review the methodology, and get the results. The execution is fully automated, but the direction is fully yours.
Deep Research runs the AI's best analysis immediately without showing you the plan first. It is faster and requires less interaction. Use it when you trust the AI to pick the right approach.
Build Mode shows the plan and lets you modify it before execution. Use it when the question is complex enough that methodology matters, or when you want to learn from the analytical process itself.
Here is every quantitative tool available through the AI Chat, with the plain-English question that triggers each one:
| Tool |
Trigger Question |
Credits |
| Correlation Analysis |
"Is X correlated with Y?" |
5 |
| Granger Causality |
"Does X predict Y?" |
10 |
| Cointegration |
"Do these two assets move together long-term?" |
10 |
| Regime Detection |
"What market regime are we in?" |
5 |
| Multi-Factor Regression |
"What factors explain my returns?" |
10 |
| Signal Decay |
"How long does this signal stay relevant?" |
10 |
| Anomaly Detection |
"Is anything unusual happening?" |
5 |
| Signal Explanation |
"Why did this signal trigger?" |
10 |
| Feature Discovery |
"Which metrics predict price best?" |
15 |
| Signal Scoring |
"How strong is this signal right now?" |
10 |
| Ensemble Builder |
"Combine my top signals into one score" |
15 |
| Alpha Leak Detection |
"Is my edge still working?" |
5 |
| Report Generation |
"Generate a full analysis report" |
25 |
| Backtesting |
"Backtest this strategy" |
30+ |
| Trade Analysis |
"Analyze my trading performance" |
10 |
You never need to remember tool names or parameters. The AI maps your question to the right tool automatically. If your question spans multiple tools, the AI chains them together. "Find what predicts my returns and then backtest a strategy based on the top predictor" triggers Feature Discovery followed by Backtest.
The Workbench provides suggested prompts organized by category. Here are the most powerful ones and what they produce.
"What's ripping right now?" → Scans real-time market data for assets with the strongest momentum across price, volume, and social metrics. Returns a ranked list with context on whether the move is backed by genuine on-chain conviction or just speculative froth.
"Top gainers with conviction" → Filters the top gainers for those backed by rising on-chain activity, exchange outflows, and smart money accumulation. Separates real momentum from pump-and-dump price action.
"Find my alpha" → Comprehensive performance attribution across regime, time, setup type, signal correlation, and emotional bias dimensions. The most powerful single prompt for understanding your trading strengths and weaknesses.
"What predicts my P&L?" → Multi-factor regression identifying which available data features have the strongest predictive relationship with your trade outcomes.
"Leading vs lagging signals" → Cross-correlation analysis identifying which signals in your watchlist lead price by the most time, giving you the earliest possible entries.
"Is my edge degrading?" → Rolling expectancy and IC analysis showing whether your strategy is maintaining its effectiveness or eroding over time.
"How should I size?" → Position sizing optimization based on your current win rate, volatility exposure, and drawdown position.
"Smart money flows" → Latest whale wallet activity, exchange reserves, and large transaction monitoring across major assets.
"Sentiment pulse" → Aggregated sentiment scores from social media, derivatives positioning, and on-chain metrics. Identifies when sentiment is at extremes that historically precede reversals.
"Divergence scan" → Finds assets where price action contradicts on-chain metrics, funding rates, or sentiment. These divergences often resolve in the direction of the non-price data.
The AI Chat is powerful, but it has boundaries. Understanding these helps you get the most from the tool.
For exploratory analysis where you do not know exactly what you are looking for. For generating initial hypotheses before diving deeper. For quick data lookups and market checks. For complex analysis that would require multiple tools and steps. For traders who are more comfortable with conversation than SQL.
The AI excels at multi-step workflows where you want answers, not code. "What happened to my win rate in the second half of last month and which assets drove the decline?" is a multi-join, multi-filter question that the AI handles in one shot. In SQL, this requires writing 20+ lines of code. In the AI Chat, it is one sentence.
For precise, repeatable queries that you run regularly. When you need exact control over joins, filters, and aggregations. When you are building a notebook pipeline where each cell feeds into the next with specific data formats. When you want deterministic results without AI interpretation.
SQL also wins for queries you plan to run many times. While the AI generates SQL each time you ask, the queries might vary slightly between runs. If you need identical output formatting every time (because a Visualization cell is linked to the SQL cell), write the SQL once and re-run it.
The AI can misinterpret ambiguous questions. "How am I doing?" could mean overall portfolio performance, recent trade performance, or performance relative to a benchmark. Be specific: "What is my trailing 30-trade win rate compared to my all-time average?"
Complex queries that involve many table joins and conditional logic may be generated imperfectly. If the AI's query does not return expected results, you can view the generated SQL, modify it manually, and re-run. The AI shows its work, so you always have visibility into the underlying analysis.
Token limits mean very long analytical pipelines may need to be broken into multiple conversations. Build Mode helps with this by structuring complex analysis into discrete steps.
The most effective Workbench users combine both methods. They use the AI Chat for exploration and discovery—"What is interesting in the data?" Then once they find something worth monitoring regularly, they take the AI-generated SQL, refine it, and save it as a dedicated SQL cell in their notebook. The AI opens the door. SQL keeps it open permanently.
This hybrid approach also accelerates the learning curve. If you want to learn SQL for crypto analysis, the AI Chat is the best teacher available. Ask a question, see the generated SQL, study how it works, and gradually start writing your own queries. Every AI response doubles as a SQL lesson tailored to your exact data and use case.
The difference between trading with and without the AI Workbench is measurable. Consider the analysis that a typical active trader needs to make informed decisions:
-
Without the Workbench: Check TradingView for price action (5 minutes). Open derivatives data platforms for funding data (3 minutes). Switch to on-chain analytics platforms for whale flows (5 minutes). Open a spreadsheet to log your trades (5 minutes). Manually calculate your win rate this month (5 minutes). Stare at the numbers and try to find a pattern (20 minutes). Total: 43 minutes of context switching, manual data gathering, and unstructured analysis. Result: a vague sense of what might work.
-
With the AI Workbench: "Give me a complete morning briefing: current regime, top funding divergences, whale activity, and my performance this week." One sentence. 30 seconds. The AI runs four analyses, synthesizes the results, and returns a structured briefing with actionable setups. You spend your time making decisions instead of gathering data.
The time savings compound over weeks and months. A trader running a 15-minute morning workflow in the Workbench does better analysis in a month than a trader spending an hour per day on manual research, because the quality and consistency of AI-powered analysis vastly exceeds what most people can produce manually across multiple tools.
This is not about replacing your judgment. It is about arming your judgment with better information, faster. The trading decisions remain yours. The analytical grunt work does not have to.
No. The AI Chat is a purpose-built system with direct access to your trading data, market databases, and 15 quantitative analysis tools. Unlike general-purpose AI, it executes real queries against real data and returns grounded results. It cannot hallucinate your win rate because it calculates it from your actual trades.
The statistical calculations are exact—they run the same algorithms that statistical software uses. The interpretations are produced by a model trained on quantitative finance and trading analysis. For standard analyses (correlation, regression, regime detection), the interpretations are highly reliable. For novel or unusual market conditions, treat interpretations as informed suggestions and apply your own judgment.
The AI provides analysis and recommendations, not trade execution. It tells you what the data shows and what historical precedent suggests. The decision to trade, and the risk management of that trade, remains yours. This is by design—automated execution without human oversight is a different product with different risk profiles.
You attach up to 3 images to a chat message. The AI's vision model analyzes the images and incorporates what it sees into its response. For charts, it identifies patterns, levels, and indicators. For trade screenshots, it analyzes entries, exits, and sizing. The images are processed securely and not stored beyond the session.
Instant (15 credits) provides quick, focused answers to straightforward questions. Think data lookups, simple calculations, and brief interpretations. Deep Research (200 credits) runs multi-step analytical pipelines that may include multiple queries, statistical tests, and visualizations woven into a comprehensive report. Use Instant for quick checks and Deep Research for serious analysis sessions.
AI Chat interactions within a notebook are saved with the notebook. You can reference previous analyses, re-run queries with updated data, and build on past conversations. Notebooks can be shared or exported so your analytical workflow is preserved.
AI Chat is available on all plans that include credits. The features are identical across plans—the difference is the number of credits allocated. Higher-tier plans provide more credits for running more analyses. Visit the pricing page for current credit allocations.
The AI accesses your trading data that has been imported or synced into Thrive. It does not connect directly to exchange APIs during analysis. You import your data through the Import cell (CSV upload or exchange sync), and the AI analyzes the imported data. This provides a security layer between the analytical tools and your exchange accounts.
The AI maintains conversation context within a session. Follow-up questions build on previous answers without needing to re-explain. "What about for ETH instead?" after a BTC analysis re-runs the same analysis for ETH. "Break that down by month" after a summary adds temporal granularity. This conversational flow makes iterative analysis natural and efficient.
Quantitative analysis is no longer locked behind a programming barrier. The Thrive AI Workbench puts 15 institutional-grade analytical tools behind a conversational interface that anyone can use. Correlation analysis, Granger causality, regime detection, Monte Carlo backtesting, trade performance attribution—all of it accessible through plain English questions.
The traders who will thrive in the next cycle are not the ones with the best instincts or the most time glued to charts. They are the ones who ask better questions of better data with better tools. The AI Workbench makes asking those questions trivially easy. The hard part is no longer the analysis. The hard part is deciding what to do with the answers.
If you have been putting off quantitative analysis because you cannot code, that excuse is gone. The tools are here. The data is live. And the AI is waiting for your first question.
→ Ask Your First Question
