Agents

The Agents page is the central hub for deploying, monitoring, and managing your AI trading agents. From here you control every aspect of how your agents interact with prediction markets — what they trade, how much they risk, and when they run. Every trade flows through a transparent pipeline you can inspect in real time.

Agent Cards

At the top of the page, a horizontally scrollable strip displays all available agents. Each card shows the agent name and avatar, the target exchange (Kalshi or Polymarket), strategy description, cumulative P&L, and a live status indicator. Cards for running agents display Stop and Nuke buttons; idle agents show a single Deploy button. If an agent is missing API keys, an amber warning appears below its card. Agents marked "Coming Soon" are visible but not yet deployable. Click any card to select it and load its detail panel below.

Performance Tab

The default tab surfaces the numbers that matter. A six-metric dashboard leads with P&L, win rate, trade count, average confidence, Sharpe ratio, and best-performing category. Below the metrics sits a cumulative P&L chart built from your actual trade history, followed by a two-column layout: open positions on the left (with unrealized P&L per market) and a live terminal on the right.

The terminal streams real-time WebSocket logs from the running agent — tagged as TRADE, SYSTEM, REASONING, or BLOCKED. You can pause and resume the feed without affecting the agent. Beneath these panels, a full trade history table lists every trade the agent has made, showing the market title, time, category, side (YES/NO), and whether the trade won or lost.

Settings Tab

Settings are organized into clear sections and are locked while an agent is running (an amber banner reminds you to stop the agent before editing).

Bot Profile

Read-only section showing the agent's description, strategy label, and LLM models used. This information is defined by the agent's codebase and cannot be changed from the dashboard.

Mode Toggle

Switch between Training and Live mode. Training mode simulates trades without placing real orders — ideal for evaluating strategy performance risk-free. Switching to Live requires your account to have live trading enabled and triggers a confirmation dialog. Live mode commits real capital through your production API keys.

API Key Status

Displays the configuration state of every key the agent needs — LLM provider keys plus exchange credentials (Kalshi or Polymarket). Each key shows a green "Configured" or red "Missing" indicator. If any key is missing, the agent cannot be deployed and you are directed to the Settings page to add them.

Trading Rules

Hard limits enforced by the backend rules engine before any trade is executed. Configurable parameters include max trade size, max open positions, daily loss limit, minimum confidence threshold, and max trades per day. Additional sections cover market filtering (minimum volume, max expiry window) and position sizing (Kelly multiplier, minimum position size, max position percentage). Changes auto-save automatically, or you can click Save Rules to persist immediately.

Trade Pipeline Tab

A step-by-step visual walkthrough of the agent's reasoning pipeline. Navigate nine steps using arrow buttons or the dot indicators: Overview, AI Models, Bot Settings, Market Ingestion, News & Sentiment, 5-Model Debate, Edge Filter & Sizing, Rules Engine, and Execution & Exit. Each step reveals the exact parameters, thresholds, and data sources the agent uses — from which LLM models participate in the adversarial debate to how Kelly sizing calculates position amounts.

This level of transparency means you never have to wonder why a trade was taken or skipped. The pipeline tab turns a black-box AI into an auditable decision process.

Deploying an Agent

Clicking Deploy on an idle agent opens a deployment dialog where you configure:

1. Duration — how long the agent runs (in minutes, or unlimited until manually stopped).
2. Cycle interval — how frequently the agent scans for new trades (default 5 minutes for Kalshi, 2 minutes for Polymarket).
3. Capital allocation — the maximum lifetime spend cap for this deployment.

After confirming, the agent starts immediately and you are redirected to the Trades page to monitor execution in real time. The agent card updates to show a running status with Stop and Nuke controls.

Agent Detail View

Clicking into an agent (via /agents/[id]) opens a dedicated deep-dive page. The header shows the agent name, status badge, GitHub repo link, exchange, and uptime. A hero section displays total P&L in large type with percentage return and today's P&L beneath it.

Cumulative P&L Chart

A full-width chart plots profit and loss over time with a time-range selector (1W, 1M, 3M, All). The chart is built from real trade data, not simulations.

Stats Grid

Five metric cards display Total P&L with trend, Win Rate (wins/total), Average Confidence, Trades Today, and Capital used out of allocated.

Trade History with Expandable Reasoning

Each trade row shows the market name, timestamp, side badge, size, confidence bar, P&L, and status pill (executed, paper, open, skipped, error). Clicking a row expands it to reveal the agent's reasoning summary — a one-line conclusion with confidence score. A "Show full reasoning" link opens the complete multi-model debate transcript in a monospace panel, so you can audit exactly why the agent made each decision.

Live Reasoning Stream

A sidebar panel streams real-time agent logs via WebSocket. Entries are color-coded by level — green for executed trades and passed rules checks, yellow for in-progress rule evaluation, and red for blocked trades and errors. Each line shows a timestamp, level badge, and message.

Controls

Three actions are available from both the agent cards and the detail view:

• Stop — gracefully pauses the agent. It retains its configuration and can be resumed.
• Resume — restarts a stopped agent in paper mode for safety (you can switch to live manually).
• Nuke — emergency kill switch. Force-stops the agent, deletes all stored API keys, and stops any other running agents. A confirmation dialog warns that you will need to re-enter credentials to deploy again. Use this only when you need to halt everything immediately.