Product Architecture

This page defines the system structure of Orien AI and the role of each layer.

Read it in two passes: first follow how a request moves through the platform, then look at the responsibility of each component.

Overview

Orien AI is organized so that interfaces remain narrow, execution remains shared, and strategy logic remains inspectable.

Three ideas drive the architecture:

• requests should pass through explicit runtime stages before they reach the market
• tools and strategies should be extensible without fragmenting the execution layer
• all trading surfaces should converge on one operational model

Execution path

Most requests begin in one of three surfaces: AI Chat, Integrated Agents, or lightweight endpoints such as Web / Telegram.

From there, the request moves through the same chain:

1. the interface captures intent
2. the pipeline runtime resolves strategy logic and orchestration
3. risk and backtest-aware checks narrow the set of allowed actions
4. the execution core handles data access, order routing, and wallet boundaries
5. the final action reaches an exchange or paper trading environment

This is why Orien AI can support different entry points without creating different trading systems behind them.

Three layers

Layer 1: Operator Interfaces

These are the surfaces where intent enters the system.

• AI Chat
• Integrated Agents
• Web / Telegram

They should stay thin. Their job is to capture intent and send it into the shared runtime.

Layer 2: Pipeline Agent Layer

This is the strategic center of the product.

It is where the platform becomes more than a tool library:

• Skills are loaded into runtime phases
• orchestration turns multi-step logic into repeatable execution
• model-assisted reasoning can be used without giving up runtime control
• risk checks and backtest assumptions stay close to the execution path

This layer is the main answer to early prompt-driven trading agents, which often mixed reasoning, execution, and policy into one opaque loop.

Layer 3: Agent Operating System

This is the shared execution substrate under all trading surfaces.

• Trade Kit provides unified exchange execution APIs
• Data API provides market and account context
• Wallet Layer defines the signing and custody boundary

Because these capabilities are shared, operators do not need a separate execution stack for each interface or each agent type.

Two ecosystems

The two ecosystems extend different attachment points in the system.

MCP ecosystem

MCP extends AI Chat with reusable tool servers.

It is designed for:

• tool discovery and installation
• chat-side workflow extension
• versioned and paid tool publishing

Skill ecosystem

Skill extends Pipeline Agents and Integrated Agents with strategy packages.

It is designed for:

• packaging trading logic into reusable modules
• binding strategy logic to runtime phases
• versioned and paid strategy distribution

One shared core

Everything converges on the same core capability: Agent Trading.

That core includes:

• market data access
• strategy evaluation
• risk enforcement
• order execution
• monitoring and reporting

The architecture matters because it lets Orien AI expand surfaces and ecosystems without duplicating the critical execution layer underneath.