3. System Architecture
Hypaw Terminal is built to manage real-time, context-aware strategies from end to end — starting with context ingestion via MCP, through trigger evaluation, and ending in optional execution via Hyperliquid or external endpoints like X (Twitter).
This section breaks down the core system components and how they interact.
Core Components
1. Terminal Interface
The browser-based dashboard at https://hypaw.ai/app is where users:
Choose and configure agents
Define context inputs and conditions
Connect accounts (X, wallet)
Monitor agent status and logs
Deploy or stop live agents
No code is required — all logic is managed through the UI.
2. Agent Runtime
Each agent is executed in a dedicated runtime environment, isolated per user. This environment:
Subscribes to specific MCP context feeds
Continuously evaluates the agent’s condition tree (trigger logic)
Handles execution when conditions are met
Agents are stateless by default, meaning they react only to present context and configuration. Persistent state (cooldowns, rate limits, counters) is handled by the underlying runtime.
3. MCP Feed Layer
The Model Context Protocol (MCP) powers real-time context delivery. For each agent, MCP:
Streams relevant context types (price, sentiment, wallet activity)
Normalizes all data into a documented structure
Updates the agent with fresh signals every few seconds
All context sources (social, onchain, market) are unified through MCP so agent logic stays clean and abstracted from API variability.
4. Execution Layer
If your agent is configured to take action (trading or posting), execution is routed through:
🔹 Hyperliquid (for trades)
Agents can place market or limit orders
Strategy parameters include size, pair, stop-loss, and cooldowns
Wallet integration is non-custodial — you approve access to the strategy logic and maintain full control
🔹 Twitter/X (for posts)
Agents can post templated content using live token, price, or trend data
Posts are rate-limited and authorized via OAuth
Future endpoints (Telegram bots, Discord messages, webhook-based alerts) are in progress.
Data & Action Flow
Here’s how data moves through the system:
Each step is decoupled, observable, and logged. If an agent condition is not met, it simply listens passively. Only when the trigger evaluates as true is execution fired.
Security Model
All trades and posts are approved and initiated from the user's wallet or authenticated account.
Agents operate in isolated environments; one agent cannot access another user’s data or logic.
You can pause or terminate agents at any time through the Terminal interface.
Next: 4. Using Terminal → Step-by-step instructions for configuring and deploying your first agent.
Last updated