Technical Architecture
Underlying Software
Last updated
Underlying Software
Last updated
Building a high-performance, low-latency UI for the Neurobro Agent service presents significant technical challenges.
Neurobro’s intelligence is powered by a dynamic swarm of specialized AI processes that coordinate efficiently to deliver intelligent responses.
To ensure scalability and responsiveness under high concurrency, our architecture is designed to efficiently distribute workloads, manage state, and optimize inter-service communication.
The Neurobro Terminal comprises three primary services:
Neurobro UI - A Next.js frontend handling user interactions.
Neurobro Backend - A FastAPI service responsible for authentication, user session management, and message routing.
Neurobro Agent - The core intelligence service that processes user queries and executes workflows.
The Agent service orchestrates a swarm of Nevrons, each executing specialized tasks.
A user submits a request via the UI, which is sent over HTTPS to the backend.
The backend authenticates the user and forwards the request to the RabbitMQ Task Queue.
The Agent service retrieves tasks from RabbitMQ, executes workflows within the Nevrons ecosystem (e.g., nevron7: market analysis, nevron31-33: technical analysis, nevron12-17: research), and streams responses back.
The backend receives updates and forwards them to the UI via HTTPS streaming for real-time interaction.
Message Broker: RabbitMQ
Database: PostgreSQL
Cache: Redis
Vector Store: Qdrant
Built with Next.js for server-side rendering (SSR) and performance optimization.
Establishes an HTTPS connection to the backend for real-time communication using streaming responses.
Renders task progress, streaming updates, and final outputs.
Implements wallet authentication via RainbowKit.
Securely stores JWT tokens for session management.
Developed with FastAPI for high-performance request handling.
Manages wallet authentication, JWT issuance, and session validation.
Routes user messages to the RabbitMQ Task Queue.
Listens for Agent responses via the RabbitMQ Response Queue and forwards them to the frontend.
Implements rate limiting and access control to prevent abuse.
Integrates with PostgreSQL and Redis for data storage and retrieval.
Runs as multiple containerized replicas, ensuring efficient handling of concurrent queries.
Executes workflow-based tasks by interacting with Nevrons.
Streams real-time results back via RabbitMQ.
Represent an ecosystem of AI-driven processes, each specializing in a particular task (e.g., market data analysis, scam detection, airdrop monitoring, etc.).
Integrate LLMs and external APIs.
Maintain persistent memory for long-term contextual understanding.
For more information about Nevrons please read the official documentation as well as our roadmap.
RabbitMQ distributes workloads across instances to prevent any single one from becoming overloaded.
As seen in the architecture diagram, multiple agent replicas allow the system to automatically scale during periods of high traffic.
A persistent message history enhances the user experience by reducing redundant requests. Additionally, HTTPS-based streaming updates ensure low-latency interactions.
The Neurobro Terminal delivers a highly scalable, low-latency architecture for interacting with Neurobro AI intelligence.
Its event-driven design ensures responsiveness, even under high user load.
The separation of concerns between UI, Backend, and Agent services enables independent scaling and future enhancements.