System Overview

DragonClaw is a modular runtime with five major subsystems.

Each part has a clear job. Together they form a chat-first crypto agent.

Main subsystems

• Connectors receive messages from chat platforms and clients

• Gateway handles auth, rate limits, validation, and routing

• Agent Loop builds prompts, runs the model, and executes tools

• Skills define what the agent can do

• Memory stores user facts and preferences across sessions

Architecture diagram

DragonClaw uses a simple four-layer architecture.

1. Connector layer receives messages from chat platforms, local tools, and custom clients.

2. Gateway layer enforces auth, rate limits, validation, and request routing.

3. Agent layer builds prompts, calls the model, and coordinates tool execution.

4. Execution layer supplies the agent with skills, memory, and model providers.

Data flow

1. A user sends a message through a connector or client.

2. The gateway validates the request and forwards it to the agent loop.

3. The agent loop builds context from skills, memory, and session history.

4. The LLM decides whether to answer directly or request a tool call.

5. DragonClaw executes the required tools and returns a formatted response.

Runtime layout

src/
├── core/
├── connectors/
└── skills/binance/