Operate an Agent

Agent Lifecycle

Agents have the following states:

Core Actions

From the agent detail page, you can:

Pause

Temporarily halts all trading activity. The agent pod stays alive but stops running cycles. Use this when you want to manually review performance or during high-volatility events.

Resume

Restarts trading from a paused state. The agent picks up where it left off.

Stop

Permanently stops the agent. The Kubernetes pod is terminated. You cannot resume a stopped agent — you’ll need to create a new one.

Restart

Stops and redeploys the agent with current settings. Use this after changing configuration in the Settings page.

Monitoring a Pipeline Agent

This is one of the most important operational surfaces in the product.

Autonomous agents should be inspectable end to end. Operators should be able to understand what the runtime did, why it did it, and what happened after execution.

What to Watch

Decisions Tab

Shows every cycle with:

• Action — LONG, SHORT, CLOSE, or HOLD
• Signal Strength — 0-100 confidence score
• Result — Filled, skipped, or rejected by risk controls
• Duration — Cycle latency

Click any row to inspect the full structured trace behind the cycle.

Trades Tab

All executed orders with:

• Side (BUY/SELL), price, amount, fee
• Realized PnL per trade
• Order ID and execution timestamp

Positions Tab

Current open positions showing:

• Entry price, current price
• Position size and leverage
• Unrealized PnL

Configuration Updates

Use the Settings area to update:

• cycle interval
• signal threshold
• visibility
• risk configuration

Changes that affect runtime behavior should be followed by a restart so the execution environment is refreshed.

Operational Principle

An agent is not “managed” only by starting or stopping it. Real operation means:

• deploying with a clear strategy
• validating with backtests
• observing runtime traces
• tuning controls
• re-running with tighter assumptions