4 Signals Your AI Agent Should Be Watching on Polymarket

In the last 24 hours, Depthy generated 23,463 signals across 3,430 markets out of 50,141 total active markets on Polymarket. Not all signals are equal. If you are building an AI agent that trades prediction markets, you need to know which signals carry the most alpha and how to consume them programmatically. Here are the four signal types Depthy tracks, ranked by reliability, with real examples from today's data.

1. Smart Money

When a wallet with a proven track record makes a move, that is signal. Depthy profiles 567 wallets and scores them by win rate, realized PnL, and consistency across market categories. When a high-score wallet executes a trade, we fire a SMART_MONEY signal with the full context: wallet address, trade size, market, outcome direction, and the wallet's historical stats.

Real example: Wallet 0xa3ad70 (composite score: 98.7, $5.3M realized PnL, 91.9% win rate across 1,247 resolved positions) has been systematically buying "No" across geopolitics, sports, and tech markets over the past 72 hours. When this wallet moves, the signal payload includes the wallet address, trade size in USDC, the target market, and the direction of the bet. Your agent gets all of this in a single JSON object.

Why it matters: Smart money signals have the highest signal-to-noise ratio of any signal type. These are not random traders placing speculative bets. They are wallets with thousands of trades and documented profitability over months. When a wallet with a 91% win rate and millions in realized profit enters a position, the information asymmetry is significant. Your agent should weight these signals heavily in its decision pipeline.

2. Volume Anomalies

Volume anomaly signals fire when a market's recent trade volume exceeds its rolling 24-hour baseline by 3x or more. This is the earliest indicator of breaking news, narrative shifts, and coordinated positioning. The signal catches movement before price fully adjusts, giving your agent a window to act.

Real example: A market on Fed policy showed a 5.2x volume spike — 156 trades in the last hour versus a 30-trade hourly average over the prior 24 hours. The spike preceded a 12-cent price move as the market repriced around new information. Agents subscribed to the volume anomaly stream caught this within seconds of the spike crossing the 3x threshold.

Depthy monitors all 50K+ active markets continuously. Volume baselines recalculate every collection cycle using a rolling 24-hour window, so you get statistically meaningful spikes, not noise from low-liquidity markets with erratic activity. The signal metadata includes the exact ratio, the raw hourly count, the baseline average, and the market context so your agent can apply its own severity filter on top.

3. Whale Movements

Whale signals track top holder position shifts across markets. When a wallet ranked in the top 20 by total position value changes their exposure significantly — adding to a position, reducing it, or exiting entirely — Depthy generates a WHALE signal. These are not just large trades; they represent conviction changes from the wallets with the most capital deployed on the platform.

Real example: A top-20 holder reducing their position in "BTC above $100K by EOY" by $8.1K, a 12% reduction in their total exposure to that market. The reduction happened across three separate transactions over 45 minutes, which Depthy aggregates into a single signal with the net change and percentage shift.

Whale movements are particularly useful for detecting early exits. When large holders start unwinding positions, it often signals a change in thesis that smaller participants have not yet priced in. Your agent can use these signals to adjust position sizing or trigger risk management logic.

4. Depth Imbalance

This is Depthy's most unique signal. We capture order book depth snapshots continuously — 1.9M+ snapshots collected to date across all monitored markets. When bid-side and ask-side liquidity diverge significantly, it indicates structural pressure that has not yet been reflected in the last trade price.

Real example: An ETH $4,400 market showed $703K stacked on the bid side versus $3.56M on the ask side. That 5:1 ask-to-bid ratio indicates substantial sell pressure waiting above the current price. Traders looking at the last trade price alone would miss that the order book is structurally tilted against further upside.

Depth imbalance is the one signal most platforms cannot replicate because it requires continuous order book collection infrastructure, not just trade data. Trade feeds tell you what happened. Depth snapshots tell you what is about to happen. For an AI agent making positioning decisions, that distinction is the difference between reactive and predictive trading.

Consuming Signals in Real Time

All four signal types are available through the REST API for polling, but the most effective integration pattern for agents is the SSE (Server-Sent Events) stream. Open a single persistent connection and receive signals as they fire, with no polling delay and no wasted requests. Here is a minimal Python example that filters for high-conviction smart money signals:

import requests
import json

API_KEY = "dpth_free_your_key_here"
url = "https://depthy.io/v1/pm/signals/stream"
headers = {"Authorization": f"Bearer {API_KEY}"}

with requests.get(url, headers=headers, stream=True) as resp:
    for line in resp.iter_lines():
        if line and line.startswith(b"data: "):
            signal = json.loads(line[6:])
            if signal["signal_type"] == "SMART_MONEY" and signal["severity"] > 80:
                print(f"High-conviction smart money: {signal['title']}")

The stream delivers all four signal types on a single connection. Filter client-side by signal_type and severity to match your agent's strategy. Free tier receives signals with a 5-minute delay; Metered and above get real-time delivery. Heartbeat pings every 30 seconds keep the connection alive through proxies and load balancers.

For agents that need historical context alongside live signals, combine the stream with the /v1/pm/signals/latest endpoint to backfill on startup. This ensures your agent has full signal history from the moment it connects, with no gaps between the initial fetch and the live stream.