Too large for one machine
Model part A
Model part B
Model part C
120B model · 3 computers · 32GB RAM each
Illustrative 120B configuration. Model fit depends on the model, quantization and available hardware. Switch computers on or off to explore.
START WITH YOUR QUESTION
Run a model you trust, choose a local AI package, or explore a new use case with us.
LOXIA GRID INFRASTRUCTURE
Loxia Grid splits a supported model across suitable computers or servers inside your organization. Run a model that is too large for a single machine by combining the capacity of your existing hardware.
We review your model, machines and network, then agree the supported setup.
Discuss your model and hardware ↗LOCAL AI PACKAGES
Choose a package built around small language models running inside your organization.
Chat with your organization’s data. Ask questions about internal files and receive answers through a local chat interface powered by a small language model.
Give your developers local AI assistance powered by a supported small language model, configured for your team’s development workflow.
USE CASE PILOT · SEPARATE SERVICE
We work with you to choose a specific task and model, agree what success looks like, and set a defined testing period. Then we run the pilot and show you the results, including what worked and where the model fell short.
A separate evaluation service to help you decide whether to proceed with a deployment.
Tell us about your pilot ↗ILLUSTRATIVE CONFIGURATION
A 120-billion-parameter model split across three computers, each with 32GB RAM and a graphics card.
Example configuration. Compatibility depends on the selected model, quantization and hardware.
HOW THE MACHINES WORK TOGETHER
Loxia distributes different parts of a supported model across connected machines. They participate in the same model execution, while the model and your data stay inside your organization.
Different machines hold different parts of the same supported model.
Loxia coordinates execution across the participating machines.
Your team uses the model through the local environment configured for your chosen solution.
FROM YOUR GOAL TO LOCAL AI
Bring your chosen model, choose a local AI package, or start with a separate pilot to evaluate a new task.
We review the model, available memory, machines and network, then agree the supported configuration.
We set up the agreed local environment and confirm the outcome with your team.
Where the configuration supports it, no new hardware is needed. If it doesn’t, we explain what must change before you commit.
BEFORE YOU START
No. The model must be supported, and the machines and network must meet its requirements. Splitting a model helps it fit across machines; it does not remove those requirements.
No. Performance depends on the model, hardware, network and workload. The illustration explains model splitting and does not represent a speed or capacity benchmark.
Yes. With a local deployment, supported AI models run on your organization’s computers and servers. Your documents, prompts and answers can stay inside your environment. Loxia Grid connects the devices that run the model. We review your chosen model, hardware, network and integrations to plan a setup that keeps processing local.
Yes. Choose the Smart Knowledge package for chat with your organization’s data, or the Software Development package for local coding assistance. Both use small language models. If you need to find out whether a model can handle a new task, start with a separate use case pilot. We agree the task, model, testing period and evaluation criteria with you before testing.
Yes. We have had good results making small language models work well in Hebrew, including with large Hebrew datasets. We can provide production-grade local AI solutions powered by small language models, with full Hebrew support inside your organization.
We can deploy Loxia Grid to a first department within 2 to 3 weeks once the scope, hardware, network access and required approvals are ready. Timing depends on your environment and coordination with your team. We confirm the deployment plan before starting.
LET’S SEE YOUR NEXT LOCAL AI DEPLOYMENT
Your chosen model. The Smart Knowledge package. The Software Development package.
Talk to Us ↗Not sure your model can do the job? Plan a pilot ↗