Skip to main content

​
Introduction

Building a high-performance, low-latency UI for the Neurobro Agent service presents significant technical challenges. Our architecture is designed to efficiently distribute workloads, manage state, and optimize inter-service communication.

​
System Architecture

The Neurobro Terminal comprises four primary services:

UI

Next.js frontend for user interactions

Backend

FastAPI service for request handling

Neurobro Agent

Core intelligence service

Signals

Orchestrator of signals generation and distribution

​
Communication Flow

0. Signal Generation

Neurobro’s signal generation pipeline:
  1. Neurobro analyzes market data and generates trading signals
  2. Signals are stored in MongoDB for persistence
  3. Backend caches signals in Redis for high-speed access
  4. UI retrieves latest signals from Backend via optimized API
  5. Users can select relevant signals to initiate targeted conversations

1. Request Initiation

User submits request via UI. Both user queries and signals are supported.

2. Authentication

Backend authenticates and forwards to RabbitMQ

3. Processing

Agent service executes workflows via Nevrons

4. Response

Real-time streaming of results back to UI

​
Core Tech

Message Broker

RabbitMQ

Relational Database

PostgreSQL

Document Database

MongoDB

Cache

Redis

Vector Store

Qdrant

SSL security

HTTPS via Caddy

​
Detailed Service Breakdown

Framework

Next.js with SSR optimization

Authentication

Web3 Auth integration

Swaps

Rango Exchange API integration
Key Features:
  • Signals Presentation
  • Chat history
  • Real-time communication
  • Task progress visualization
  • Streaming updates
  • JWT token management
  • Core Features
  • Integrations
  • FastAPI implementation
  • Authentication management
  • Message routing
  • Signals distribution
  • Rate limiting
  • Containerized replicas
  • Workflow-based execution
  • Real-time streaming
  • Nevron orchestration
There’re 50+ Nevrons in total, each with a unique purpose and set of tools. Here are some examples:

Market Analysis

nevron7

Technical Analysis

nevron31-33

Research

nevron12-17

On-Chain scam detection

nevron37

Generation

Agent service creates actionable insights by orchestrating nevrons

Storage

Saved in MongoDB

​
Scalability & Performance

Since the Neurobro Agent is a single instance, which is not scalable, we found a way to distribute the workload across multiple orchestrators of the swarm of nevrons.

Load Distribution

RabbitMQ workload balancing

Auto-scaling

Multiple agent replicas

Low Latency

RPC streaming

​
Additional Resources

Nevron Documentation

Learn more about the Nevron framework

Changelog

See what’s new in the Terminal
For technical support or architecture discussions, join our developer community.
⌘I