> ## Documentation Index
> Fetch the complete documentation index at: https://docs.engramlabs.xyz/llms.txt
> Use this file to discover all available pages before exploring further.

# Architecture

> Five layers from data ingestion to collective intelligence

## Five layers, each with a distinct responsibility

**Data Ingestion** – Background polling across 20 intelligence layers. WebSocket-first for streaming data, REST for snapshot queries and fallback.

**Intelligence Cache** –  Structured, queryable representation of current market state indexed by asset. All UI and agent queries read from cache first. Direct API requests are the last resort.

**Agent Reasoning Engine** – Reads from cache, reasons about market state, generates proposals, monitors positions, writes lessons. LLM-agnostic via BYOLLM. Skill files encode the agent's knowledge — versioned, updatable, shared across the network.

**Interaction Layer** – The terminal UI. Conversational interface where the agent panel is the primary surface. Chart and overlays provide supporting context.

**Collective Learning Layer** – Lesson library, pattern extraction, skill file updates. Individual insight becomes collective intelligence.

## Core Principle

The data stack behind each agent comprises hundreds of individually mapped endpoints across 50+ specialized providers. Every market on Hyperliquid is continuously monitored, with the fastest streams updating sub-second.