AI Chat
Ask the AI anything about markets, your portfolio, whales
AI Chat is your conversational analyst inside Predite. Instead of clicking through scanners, charts, and tables to assemble a picture of a market, you ask a question in plain English and get an answer grounded in **live market data**, **your own portfolio**, and **whale activity** that Predite is already tracking for you. It lives at `/dashboard/chat` and is built for the moments when you want a fast, contextual read — "is this priced right?", "what changed since this morning?", "how exposed am I to this event?" — without leaving the dashboard.
This page explains what AI Chat can and cannot do, how to use it well, the prompts that get the best results, plan limits, and how it connects to the rest of Predite.
## What AI Chat actually does
AI Chat is a retrieval-grounded assistant. That phrase matters: it is **not** a generic chatbot answering from memory. Before it responds, Predite pulls the relevant context and feeds it to the configured AI model so the answer reflects current reality rather than stale training data.
When you send a message, the assistant can draw on:
- **Live market data** from Polymarket and Kalshi — current YES/NO prices, bid/ask spread, recent CLOB order-book depth, 24h volume, liquidity, and resolution criteria for the markets you reference.
- •**Predite's AI probability model** — the platform's independent fair-value estimate for a market, so the assistant can compare the traded price against a modeled probability and talk about edge in EV terms.
- •**Your portfolio** — open positions, average entry, unrealized and realized P&L, and concentration across correlated markets, so answers about exposure and sizing are specific to *you*, not generic.
- •**Whale activity** — recent large orders and notable wallet flows surfaced by the Whale Tracker, so the assistant can tell you when smart money is moving into or out of a market you ask about.
- •**News and the event calendar** — upcoming catalysts and recent headlines tied to a market, so it can flag *why* a price may be moving.
The assistant stitches these sources together. Ask "fair price for the Fed cut market?" and it can quote the current Polymarket price, Predite's modeled probability, the implied edge, and whether whales have been accumulating — in one answer, with the numbers attached.
### What it is **not**
AI Chat is an analysis tool, not an advisor and not an execution engine.
- **It is not financial advice.** Every output is informational. The assistant does not know your full financial situation, risk tolerance, or tax position, and prediction markets carry real risk of total loss. Treat its answers as a research starting point, verify the underlying numbers, and make your own decisions. Nothing it says is a recommendation to buy, sell, or hold.
- •**It does not place trades.** AI Chat cannot submit orders, cancel orders, or move funds. Placing live orders is handled separately and requires the **Bot plan** plus a connected wallet (see Live Trading). Chat can help you *reason about* a trade; it cannot execute one.
- •**It can be wrong.** Models can misread context or hallucinate a number that "looks right." When a figure is load-bearing for a decision — a price, a spread, an edge — confirm it against the live market page or the EV Scanner before acting.
## Getting started
1. Open **`/dashboard/chat`** from the left navigation, or click the chat icon available on most market and portfolio views. 2. You'll land on an empty thread with a row of **suggested prompts** above the input box. These are a fast way to see the feature work — tap one to send it immediately. 3. Type a question in the input box and press **Enter** (use **Shift+Enter** for a multi-line message). The assistant streams its reply token by token, so you can start reading before it finishes. 4. **Reference specific markets** by name, by Polymarket/Kalshi slug, or by pasting a market URL. The more precisely you point at a market, the better the grounding — vague references force the assistant to guess which market you mean. 5. Ask **follow-ups in the same thread**. Context carries forward, so you can drill in: "now compare that to the Kalshi equivalent," then "which has the better spread for a $2k position?" 6. Start a **new chat** when you switch topics. A fresh thread keeps context clean and avoids the model dragging an earlier, unrelated market into a new answer.
### Reading the response
Answers typically combine a short plain-English read with the data behind it. When the assistant cites a price, an edge, or a whale flow, it's pulling from the same live sources that power the rest of Predite — so the numbers should line up with what you see on the market page. If something looks off, that's your cue to verify before trusting it.
## Example questions that work well
AI Chat rewards specific, well-scoped questions. Some patterns that consistently produce strong answers:
### Fair-value and edge checks
- *"What's a fair price for [market]?"* — returns the current traded price, Predite's modeled probability, and the implied edge.
- •*"Is [market] overpriced right now?"* — frames the gap between price and model in EV terms.
- •*"Where's the EV on the YES side of [market] at the current ask?"* — ties the question to the actual price you'd pay, including the spread.
### Head-to-head comparisons
- *"Compare these two markets: [A] and [B]."* — lines up price, liquidity, spread, volume, and modeled probability side by side.
- •*"Polymarket vs Kalshi for [event] — which has the better entry?"* — useful for spotting cross-venue pricing gaps. For acting on persistent gaps, see the Arbitrage tool.
### Portfolio-aware questions
- *"How exposed am I to the election outcome across all my positions?"* — aggregates correlated exposure, not just a single ticker.
- •*"What's my current unrealized P&L and which position is the biggest drag?"*
- •*"If [event] resolves NO, what happens to my book?"* — a quick scenario read across open positions.
### Context and catalysts
- *"Why did [market] move today?"* — pulls recent news and calendar events tied to the market.
- •*"Are whales buying or selling [market] this week?"* — summarizes large-order flow from the Whale Tracker.
- •*"What catalysts are coming up for [market]?"* — surfaces scheduled events from the News & Calendar.
### Suggested prompts
The chips above the input box are dynamic — they adapt to what you've been looking at and to active opportunities on your dashboard. If you have an open position in a market that just saw a large whale order, a suggested prompt may offer to summarize it. Treat the chips as shortcuts to the most relevant question right now, not a fixed menu.
## Tips for getting better answers
- **Name the market precisely.** "The CPI market" is ambiguous when several are live. Paste the URL or use the exact title and the assistant grounds on the right one.
- •**Ask for the math.** Add "show the EV" or "walk through the calc" and you'll get the reasoning, not just a verdict — which also makes it easier to spot when a number is wrong.
- •**One question at a time for hard asks.** A focused prompt gets a focused, accurate answer. Bundle five questions into one message and quality drops.
- •**Use follow-ups instead of restating.** The thread remembers. "Now size that with the Kelly Calculator at half-Kelly" builds on the prior answer cleanly.
- •**Cross-check load-bearing numbers.** For anything you'll trade on, confirm the price and spread on the live market page or in the EV Scanner. Chat is a fast first pass, not the system of record.
- •**Lead with your constraint.** "For a $500 position with low slippage tolerance, which of these two is better?" gives the assistant the frame it needs to actually be useful.
## Plan gating and rate limits
AI Chat is available on every paid plan, but capacity scales with your tier.
- **Starter ($29/mo)** — AI Chat with grounding on live market data and Predite's AI probability model, at the Starter message allowance. Best for occasional fair-value and comparison checks.
- •**Pro ($59/mo)** — a higher message allowance plus deeper grounding, including richer portfolio context and full Whale Tracker integration in answers. The right tier if chat is part of your daily workflow.
- •**Bot ($99/mo)** — the highest message allowance and the most context, alongside the bot and live-trading features. Note that **chat itself never places trades** — execution is a separate Bot-plan capability. Chat complements it by helping you reason about setups the bot or you will act on.
### How rate limits work
- Usage is metered as **messages per period** (rolling window) tied to your plan, not by counting individual words or tokens.
- •When you approach your limit, the input box shows your remaining allowance so there are no surprises mid-thought.
- •If you hit the limit, sending pauses until the window resets, or you can upgrade for immediate additional capacity.
- •Limits exist to keep responses fast for everyone and to control the cost of the underlying model — heavy, automated workloads belong in the bot and API features, not in interactive chat.
### Which model answers you
AI Chat runs on the **AI model configured for the platform**. Predite selects and maintains the model so you get a consistent, capable assistant without having to manage keys or pick a provider. Because grounding (live prices, your portfolio, whale flow) is layered on top of the model, the *freshness and accuracy of the data* comes from Predite's pipelines — the model's job is to reason over that context and explain it clearly. If the configured model is updated, your experience improves automatically; you don't need to change anything.
## How AI Chat fits the rest of Predite
AI Chat is most powerful as connective tissue between Predite's analytical tools. A typical flow:
1. The **EV Scanner** flags a market trading away from its modeled fair value. 2. You open **AI Chat** and ask "why is the EV Scanner flagging [market]?" to get the context — model probability, spread, recent news. 3. You ask "are whales in this?" to confirm whether smart money agrees, pulling from the **Whale Tracker**. 4. You size the idea with the **Kelly Calculator**, then test it risk-free in **Paper Trading** before committing capital. 5. If you decide to act, you place the order yourself or via the **bot** on the **Bot plan** with a connected wallet — chat does not execute for you.
Because the assistant reads the same live data and the same AI probability model that power those features, its answers stay consistent with what you see elsewhere on the dashboard. It is the fastest way to *interrogate* the signals Predite generates — and a reminder, every time, that the final call is yours.
## Related Docs
- [Quickstart](/docs/quickstart)
- •[EV Scanner](/docs/ev-scanner)
- •[AI Probability Model](/docs/ai-probability)
- •[Whale Tracker](/docs/whale-tracker)