Dashboard
Terminal
Live execution feed and signal monitoring.
The Terminal is your real-time command center for monitoring every decision your AI agents make. Every market scan, every AI debate, every risk-rule check, and every order execution streams live to this page — giving you full transparency into autonomous trading as it happens.
Overview
The Terminal page consolidates two critical data streams into a single view: raw execution logs from your running agents, and a structured signal pipeline that visualises how each market opportunity progresses from initial scan to final execution (or rejection). An action bar at the top shows how many agents are currently active and provides emergency controls — Stop All and Nuke — for immediate intervention.
Two-Panel Layout
The Terminal uses a split-screen layout on desktop, stacking vertically on mobile. Each panel has its own agent filter dropdown so you can isolate output from a single agent or view all agents simultaneously.
Left Panel — Terminal Logs
The left panel is a live, streaming console powered by a persistent WebSocket connection to your backend. Every meaningful log line from every running agent appears here in reverse-chronological order, with colour-coded type badges for instant recognition.
- Logs are delivered in real time over WebSocket — no polling, no page refresh required.
- A noise-suppression filter automatically hides low-level HTTP traffic, SDK initialisation chatter, and internal API calls, surfacing only the trading-relevant output you care about.
- Errors and failures always bypass the noise filter — they are never hidden.
- Auto-scroll keeps you pinned to the latest entry. Scroll up manually to pause, and the feed holds position until you return to the top.
- Use the agent dropdown to isolate logs from a single agent, or select "All Agents" for the full firehose.
Right Panel — Signals
The right panel shows the signal pipeline: the structured journey each market opportunity takes through your agent's decision process. Signals flow through six stages — Scan, Filter, Debate, Rules, Queue, Execute — and the panel offers two ways to watch them.
- Cards view — Groups signals by pipeline stage with count badges, colour-coded headers, and clickable cards that open a detail modal with full reasoning.
- Track view — A canvas-based race-track animation where each signal appears as a dot travelling left-to-right across stage markers. Killed signals show a red X at the stage they were rejected; executed signals reach the green finish line with a checkmark.
Toggle between views using the Cards / Track switch in the panel header. Both views share the same agent filter, so you can drill into a single agent's pipeline at any time.
The Floating Terminal
When you navigate away from the Terminal page while agents are running, a floating mini-terminal appears in the bottom-right corner of every page in the dashboard. It keeps you connected to live execution without leaving your current workflow.
Minimised State
The minimised state is a compact black pill labelled "Agent Running" with a pulsing green dot. Click it to expand the terminal overlay.
Expanded State
The expanded overlay shows the last 50 meaningful log lines in a scrollable, 240px-tall panel with the same colour-coded log levels as the full Terminal. A header displays a live log count and two controls: an expand button that navigates to the full Trades page, and a close button that minimises the terminal back to a pill.
Behaviour
- Auto-opens the moment a WebSocket message arrives from a running agent — you never need to open it manually.
- Applies the same noise-suppression filter as the main Terminal panel, hiding HTTP chatter and SDK internals.
- Auto-scrolls to the latest log entry so the most recent activity is always visible.
- Disappears entirely when no agents are running and no messages are buffered.
Understanding Terminal Output
Every log line is tagged with a colour-coded type badge. The table below shows the categories you will encounter and what they mean.
| Badge | Colour | Meaning |
|---|---|---|
| LOG | Grey | General informational output from the agent process. |
| SCAN | Blue | Market discovery — fetching eligible markets, upserting data. |
| AI | Orange | AI model inference — analysis, ensemble predictions, probability estimates. |
| RULE | Amber | Risk-rule evaluation in progress (edge checks, position limits, cash reserves). |
| RULES ✓ | Green | All risk rules passed — the signal is cleared for execution. |
| REJECT | Red | A risk rule blocked or rejected the trade. |
| TRADE | Green | Order execution — paper or live trade placed on the exchange. |
| ALLOC | Purple | Portfolio allocation — Kelly sizing, capital distribution, Sharpe optimisation. |
| RISK | Pink | Risk monitoring — drawdown alerts, rebalancing, volatility checks. |
| ERROR | Red | An error or failure that needs attention. Errors are never filtered out. |
Example Log Patterns
A typical agent cycle produces logs in this sequence, giving you a narrative of the agent's reasoning from start to finish:
- Market scan — "Fetched 142 eligible markets" / "12 markets to process" (SCAN badge).
- AI analysis — "Analyzing market: Will X happen?" / "predicted_prob=0.72" (AI badge).
- Risk-rule checks — "Checking rules..." / "EDGE APPROVED" / "POSITION LIMITS OK" / "CASH RESERVES OK" (RULE/RULES badges).
- Execution — "Executed position: YES @ 0.45, size $25.00" (TRADE badge).
If a signal is rejected at any stage, you will see a REJECT badge with the specific rule that blocked it — for example, "EDGE REJECTED" or "CASH RESERVES INSUFFICIENT".
Signal Views
Cards View
The Cards view organises active signals into collapsible stage groups. Each stage header shows its label and a live count. Individual signal cards display the market ticker, and clicking a card opens a detail modal with the full reasoning chain, AI probabilities, and rule verdicts for that signal. A stats bar at the top provides an at-a-glance summary including the number of signals killed at each stage.
Track View
The Track view renders a canvas-based race-track visualisation. Each signal is a coloured dot that animates from left to right across vertical stage markers (Scan, Filter, Debate, Rules, Queue, Execute). Up to 10 signals display simultaneously in parallel lanes. Killed signals stop with a red X at the stage where they were rejected and fade out. Signals that reach execution hit the green finish line with a checkmark. The animation runs at 60fps using requestAnimationFrame for smooth, GPU-friendly rendering.
Action Bar
Above the two panels, an action bar displays the current agent status (e.g., "2 agents running") with a pulsing green indicator. When agents are active, two emergency controls appear:
- Stop All — Gracefully stops all running agents and cancels pending orders. You can redeploy from the Strategies page afterward.
- Nuke — Force-kills all agents, cancels orders, and deletes all stored API keys. This is a last-resort kill switch; you will need to re-enter exchange credentials to deploy again.
Both actions require confirmation before executing. A link to the Audit Logs page is also available for reviewing historical execution records.
Use the Terminal to debug rejected trades. When a signal is blocked, the logs show exactly which rule fired — edge threshold, position limit, or cash reserve — so you can adjust your strategy parameters and redeploy with confidence.