# Core Architecture

{% stepper %}
{% step %}

### The Coordination Layer (Orchestrator Agent)

* Sitting at the heart of the swarm, the **Neko Orchestrator Agent** acts as the central brain. It receives parsed user intent from the Interaction Layer and formulates a strategic plan for execution.
* It coordinates tasks, manages state, and delegates specific actions to specialized agents within the Execution Layer. It also interfaces with external partner APIs like **Hypurr Bot** or **pvp.trade** where necessary.
  {% endstep %}

{% step %}

### The Intelligence Layer (AI Core)

This layer provides the cognitive power enabling Neko’s intelligent decision-making. It comprises:

* **Advanced Context Engineering:** the multi-agent system gets both historical and real-time market data from third-party protocols via a combination of proprietary subgraphs, APIs, and the cadence mechanism. This allows the agent to execute upon financial alpha as actionable signals populate into the context window of the multi-agent system.
* **AI Memory & Retrieval (RAG/CAG):** Neko utilizes **Retrieval-Augmented Generation (RAG)** to fetch real-time data, contextual information, and semantic knowledge from vector databases (Vector DB / Semantic Memory). **Cache-Augmented Generation (CAG)** leverages heuristics and frequently accessed data (like agent transaction history or optimized parameters) for efficient decision-making. This allows agents to recall past actions and learn effective strategies. *(Refer to Image 2 for AI Agent System detail).*
* **Proprietary Models:** Machine Learning models (e.g., **LSTM, ARIMA, Prophet, GNNs**) analyze market data, predict trends, assess risk, and optimize strategies.
  * Future Development
* **Reinforcement Learning (RL):** Models are continuously refined through feedback loops (potentially using techniques like DPO), allowing the system to adapt and improve its execution and risk management over time.
  * Future Development
    {% endstep %}

{% step %}

### The Execution Layer (DeFi Multiagent System & Contracts)

* This layer translates the Orchestrator’s plan into on-chain actions.
* **Specialized Agents (Architect Agents / RIG Agents / Neko Bots):** These agents possess specific capabilities, such as executing trades via optimal routing (Shogun architecture), managing assets within yield vaults (**Neko Vaults**), handling complex collateralization for perpetuals, or interacting directly with specific HyperEVM protocols (**LST, Lending, CDP, AMMs**) and the Hyperliquid L1 orderbook via their respective APIs. They employ multi-step workflows involving planning and tool-calling.
* **Smart Contracts:** Secure and audited smart contracts form the bedrock for holding user assets and executing specific, predefined DeFi logic on HyperEVM.
  * The `Vault` & `VaultFactory` contracts are derived from Morpho Vault V2 and are identical in the implementation
  * `UniversalAdapter` & `UniversalAdapterFactory` are custom contracts to enhance programability and flexibility of the vault contracts. Read (audits) to read the security reviews.
    {% endstep %}

{% step %}

### The Interaction Layer (Social Agents) — Future Development

* This is the primary user touchpoint. Users interact with Neko via simple commands or natural language mentions on familiar social platforms like **X, Discord, and Telegram**.
* This social agent is an Eliza Agent and will serve as an alpha distillation agent (think AIXBT for Hyperliquid) as well as an ad network.

Example social action:

* Compose this tweet: @neko\_hl gneko — this will send a link to the application for users to start trading
  {% endstep %}
  {% endstepper %}


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://docs.neko.fun/architecture/core-architecture.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.