Deploy an Agent

This guide covers the current primary operating path in Orien AI: deploying a Pipeline Agent on top of the shared trading core.

The goal is not only to launch a runtime, but to launch one that can be understood, monitored, and adjusted after deployment.

Before you start

Before you create an agent, confirm four things:

• you have selected a Skill that matches the market you want to trade
• the required secrets or exchange credentials are available
• you know whether the first run should be Paper, Backtest, or Live
• you have a clear first-pass risk configuration

If any of these are missing, deployment usually succeeds technically but fails operationally.

1. Choose a Skill

Navigate to the Skills Market  and select the strategy package you want to run.

The Skill defines the operating contract for the agent:

• supported pairs or markets
• expected runtime parameters
• strategy logic and tool dependencies
• the runtime phases that participate in each cycle

If the Skill is unclear about markets, runtime parameters, or execution assumptions, stop there and resolve that ambiguity before deploying.

2. Configure the runtime

Basic information

Market and pair selection

Choose one of the pairs supported by the Skill, or enter a custom pair when the Skill allows it.

Good practice here is simple: match the pair to the strategy package, not to operator preference. A mismatch between strategy assumptions and market selection is one of the most common causes of poor results.

Trading mode

Runtime parameters

Typical parameters include:

• Cycle interval for how often the agent runs
• Signal threshold for how much confidence is required before action
• LLM model when the Skill contains model-assisted reasoning steps

Start with the Skill author’s defaults when they exist. Change one parameter at a time when tuning behavior.

3. Set risk boundaries

Risk controls are part of the runtime contract, not an optional layer added after the strategy is written.

At minimum, review:

• position limits
• leverage limits
• drawdown limits
• stop loss and take profit settings
• trade frequency constraints

See Risk Controls for the full model.

4. Run a backtest first

If the strategy is new, modified, or unfamiliar, run a backtest before live deployment.

Backtesting helps you answer four basic questions:

• does the strategy produce signals in the expected market conditions
• do the risk limits block actions that should remain valid
• are trade frequency and sizing consistent with the intended style
• is the agent behavior understandable enough to monitor after launch

Backtesting does not remove live risk. It reduces avoidable ambiguity before the first live cycle.

5. Deploy to paper or live

When you deploy, the platform will:

1. create the runtime record
2. provision the execution environment
3. download and load the Skill package
4. initialize the trading runtime
5. start the cycle loop or backtest replay

For a first deployment, Paper is usually the correct starting point even if the long-term goal is live trading.

6. Monitor the first cycles

Deployment is only the beginning. The first few cycles tell you whether the runtime is behaving the way the strategy description suggested.

Inspect:

• cycle decisions
• trade execution results
• positions and exposure
• backtest outputs when applicable
• runtime health and failure states

Common failure cases