The BitFlux Solution

What is Bitflux

The goal of BitFlux is to let you squeeze more value out of your compute resources. The BitFlux solution provides the following benefits:

  • Actively improves EC2 instances efficiency
    • Smaller or fewer instances
    • Reduce your compute costs by up to 50%
  • Accurate sizing of EC2 instances
  • Better insight into your workload resource usage

Basic Solution Components

AMI for AWS EC2 instances

The AMI is derived from the Ubuntu Server 24.04 LTS AMI modified to include:

  • A customized Linux kernel
  • The bitfluxd system service

The bitfluxd daemon runs as service using a high performance custom AI model to predict which memory isn't needed then instructs the custom kernel to proactively free it.

MCP server

To help the user interact with the instance for monitoring purposes the Model Contect Protocol server, bitflux_mcp, runs on the users local workstation and interacts with an AI agent to aid in querying the instance for memory usage statistics through the cloud service.

Cloud service

The cloud service is a gateway for the APIs the MCP server uses to interact with BitFlux on each instance. Using the MCP server your AI agent can identify your EC2 instances running bitfluxd and query them for memory usage statistics directly in realtime.

Quick Start

Evaluation

BitFlux has a free solution to evaluate existing workloads running on Ubuntu EC2 instances with the goal of predicting the best EC2 instance size to use with BitFlux. Using the bitflux_mcp server most AI agents can run a controller experiment with some scaffolding see: https://github.com/bitflux-ai/examples

Install the BitFlux Evaluator

Running the following on a AWS EC2 instance running Ubuntu 24.04 will install a service called bitfluxeval using a .deb package.

curl --proto '=https' --tlsv1.2 -sSf https://apt.bitflux.ai/install_eval.sh | bash

The bitfluxeval service will monitor your system and register with the BitFlux cloud service. To uninstall simply run

sudo apt remove bitfluxeval

Install the BitFlux MCP server

The MCP server is installed on your local workstation in one of several ways. bitflux_mcp Note that some features require you have aws tokens configured to interact with AWS.

See https://github.com/bitflux-ai/bitflux_mcp for details.

Using the MCP server with an AI agent

AI agents can use bitflux_mcp to list available EC2 instances running a BitFlux AMI or the bitfluxeval service, query them for memory usage statistics, and make recommendations for instance size.

AWS EC2 AMI Usage Instructions

We offer a turnkey solution in the form of a AMI on the AWS Marketplace. Search for "BitFlux" in EC2 -> AMI Catalog -> AWS Marketplace AMIs

Launch the AMI:

  • Select the BitFlux AMI from AWS Marketplace and choose a compatible instance type (e.g., t3a.nano, vendor recommended).
  • Configure the instance as you would a standard Ubuntu Server 24.04 LTS AMI.

Connect to the Instance:

  • Default SSH username: ubuntu.
  • Use SSH to connect: ssh ubuntu@<instance-public-ip>.
  • Ensure your key pair is configured during launch.

Security Group Configuration:

  • Inbound: Open TCP port 22 (SSH) to 0.0.0.0/0 for SSH access. For security, restrict to your trusted IP range (e.g., 192.168.1.0/24).
  • Outbound: At a minimum, allow TCP port 8883 (MQTT over TLS) to 0.0.0.0/0 to enable bitfluxd to connect to the BitFlux cloud service via AWS IoT MQTT. If cloud connectivity is disabled (see below), this is not required.
  • Note: These settings align with standard Ubuntu AMIs, with port 8883 added for BitFlux.

Configure and Use BitFlux Features:

  • The bitfluxd daemon is pre-installed and runs automatically, optimizing RAM usage (30–50% savings).
  • Check bitfluxd status: systemctl status bitfluxd.
  • View logs: journalctl -u bitfluxd.
  • To disable cloud connectivity (e.g., no telemetry), create /etc/bitflux/config/.noconnect: sudo touch /etc/bitflux/config/.noconnect.

Monitoring and Health Checks:

  • Monitor RAM usage: cat /proc/meminfo or use free -m.
  • Verify bitfluxd is active: systemctl is-active bitfluxd.
  • Use the BitFlux MCP server https://github.com/bitflux-ai/bitflux_mcp to query telemetry (e.g., memory stats) via the BitFlux cloud service.

Post-Launch Configuration:

  • To update bitfluxd and the bitflux kernel use apt as normal: sudo apt update && sudo apt upgrade.
  • For advanced monitoring, integrate with your DevOps tools using the MCP server.

Open Source Code

Open Sourced patches and build scripts used to create the custom kernel build can be found at https://github.com/bitflux-ai/bitflux_kernel