🏷️ €19900 | Base Price

MX-AI License including 1 year of updates and support 🔹

+€4400 for 1 additional year of updates and support 🔹

+€9900 for 3 additional year of updates and support 🔹

Contact us Contact us for academic pricing or tailored packages.

MX-AI is the BubbleRAN automation and intelligence platform that bringing autonomous operations and optimization to 5G/6G networks from concept to realization.. It implements both AI-for-RAN and AI-on-RAN Apps in a form of Multi-Agent Workflows and Blueprints, orchestrating collaborative AI agents to enable closed-loop management and control across the full lifecycle (Day 0/1/2+). These workflows and blueprints transform network observability into insights and insights into coordinated actions across time and space.

The platform is powered by the BubbleRAN Agentic Toolkit (BAT), providing a ready-to-use AI-RAN catalog of reusable and extensible agents, workflows, and blueprints, accessible via TelcoFabric Portal.

Proven in real deployments and adopted by 50+ partners worldwide, MX-AI is a leading platform for AI-driven 5G/6G O-RAN, enabling the realization of AI-RAN workflows and blueprints, and real-time network optimization at scale.

📄 MX-AI Data Sheet | 💡 MX-AI Walkthrough | ▶️ MX-AI Webinar

💡 Getting started is simple: choose or customize an AIFabric describing your Blueprint and Workflow from the TELCOfabric portal, and deploy it with a our GUI or command line:

#Deploy your multi-agent AI fabric in a single command
bash$ brc install aifabric myfabric.yaml

What’s Included?

1. Multi-Agent AIFabric: Cloud Native multi-agent runtime, Agent-to-Agent communication with A2A, exposure of algorithms, datasets and ML models with MCP
2. Agents Catalog: Supervisor Agent, SMO Agent, API Agent, SLA Agent (PoC), UL Anomaly Detection Agent (PoC)
3. BubbleRAN Agentic Toolkit (BAT): Agents Development Kit (ADK), CLI tool to create, build, benchmark and publish agents
4. TelcoFabric Access: Pull/Push Agents, LLMs, ML models, Datasets and other artifacts (with versioning)
5. Web GUI, CLI & Chat: User-friendly UI for chatting with the Agents
6. Training Materials: Access to Lab instructions, code samples, and tutorials.

🚀 Add-on Specialized Agent

Tailored AI-agent development is available as an add-on service. Contact us for pricing and delivery of a personalized agentic solution.

Below is a summarized overview of our Agents and MCP Servers Catalog. For more details, visit our Open Documentation.

🟢 AVAILABLE 🟣 EXPERIMENTAL 🟡 PLANNED

Multi-Model Support

MX-AI provides support for several language models providers.

• We recommend a minimum of 16 GB of GPU memory to deploy local SLMs/LLMs. High-scale deployment may need A100/H100 GPUs.
• Logos are trademarks of their respective owners and used here for identification only.
• Support for the following model providers has been tested:
  • OpenAI
  • Nvidia
  • Ollama
• Support for the following model providers is offered but not fully tested yet:
  • Anthropic
  • Deepseek

Multi-Agent AIFabric Example

|

Mx-AI Benefits at a Glance

(*) Digital Twin functionalities are provided by MX-DT, which is a separate product and is not included by default in the MX-AI package.

Practical Use-Cases

Optimize RAN Configuration

• Observability rApp: collect RAN KPIs.
• ML Model: use KPIs and current configuration to predict a tentative improved configuration.
• Configuration Agent: access the rApp and ML Model with MCP to satisfy user requests. Enforce new configurations.
• Example query: Propose a new P0 nominal value to maximize the uplink throughput of the RAN node gnb1.

SLA Enforcement

• SLA rApp/xApp: controls the RAN PRB in a closed-loop fashion to create a network slice offering a certain DL/UL throughput.
• SLA Agent: access the SLA rApp with MCP to translate high-level intents into requests for the SLA rApp.
• Example query: I want to use the network ‘sample-network’ to watch videos in 4K and download large files.

Uplink Anomaly Detection

• Observability rApp: collect RAN KPIs.
• ML Model: use KPIs time-series to detect interference in uplink. Generate anomaly reports.
• UL AD Agent: retrieve anomaly reports and provide a tentative root-cause analysis.
• Example query: Did you detect any malfunctioning between 3AM and 5AM?.

Ready to test?

bash$ brc install aifabric myfabric.yaml

• BubbleRAN Agentic Toolkit
• MX-AI Agents Training - Tutorial in our Open Documentation
• BubbleRAN Command Line (brc) - Tutorial in our Open Documentation
• Would you like to talk with our team? Book a live demo

Need more information?

Check our frequently asked questions about MX-AI to learn more and get quick replies.

BubbleRAN also offers tailored AI-agent development as an add-on service. Whether you need a specialized agent or want to extend the MX-AI platform with custom capabilities, our team can work with you to define the technical specifications and deliver a personalized solution. Contact us to discuss your needs and explore the details.

To answer your unique deployment and projects needs, we can plan a live demo, help you forward with a requirements questionnaire, and connect you with our partner ecosystem (universities, system integrators, cloud providers). 📧 contact@bubbleran.com

Frequently Asked Questions

1️⃣ Can MX-AI run fully on-prem?

Yes. SLM-first deployments run on modest edge GPUs. You can burst to larger GPUs when needed.

2️⃣ Do you integrate with our SMO/RIC?

MX-AI is O-RAN-friendly (R1/A1) and offers REST/gRPC adapters for common SMO/RIC NBI.

3️⃣ How do we publish and reuse agents?

Via TelcoFabric. Pull/push agents, datasets, and tuned models with versioning and fork/merge workflows.

4️⃣ What hardware do we need?

MX-AI runs great on a single 16–24 GB GPU.
For training or large-scale inference, use A100/H100 class GPUs.

5️⃣ What if I don’t have my own GPU ?

You need to purchase an API key from an external provider, as this is not included by default.

6️⃣ How is data privacy handled?

Data stays local in sovereign mode.
We support audit logging, role-based access, and optional federated learning.
No telemetry leaves your site unless you opt-in.

7️⃣ Why MX-AI ?

• Tame 5G/6G complexity - Autonomously orchestrate multi-vendor RAN/Core with closed loops.
• Proactive operations - Agents detect anomalies, predict faults, and cut MTTR from hours to minutes.
• Sovereign by design - Run SLMs (5–20 GB) fully on-prem/edge; burst to large GPUs only when needed.
• Digital-twin safeguard - Test policies in MX-DT and push only if better (“apply-if-better” gates).
• One hub for reuse - Share xApps/rApps, agents, and tuned models via TelcoFabric, no lock-in.

8️⃣ Ask your questions