Data stays in your network
Prompts, documents and answers are processed on servers you own, inside your own data centre.
Cortex On-Premise
Size your private AI deployment.
Run TextCortex on servers in your own data centre. Set your headcount, usage and open-weight model, and see the GPU servers, load and electricity cost it takes.
Estimates based on measured TextCortex enterprise usage. Ctrl/⌘ + scroll to zoom, drag to move.
Why on-premise
Your models, your hardware, your network.
Open-weight models you choose
Compare DeepSeek, GLM, MiMo, Kimi, Gemma, Qwen and Nemotron on quality, memory and speed before you buy hardware.
Costs you can plan
See the server tier, the number of nodes and the electricity cost for your country before you ask for a quotation.
How the sizing works
Sized for your busiest hour, not an average day.
Measured demand
Usage profiles come from real enterprise use of TextCortex, from occasional chat to agent-first power users working on long documents.
Peak load
Servers are sized for five times the busiest-hour average, and never below the burst of a single power user.
Clearly marked estimates
Per-server throughput is either measured in public benchmarks or marked as an estimate. Final sizing is confirmed with a load test.
FAQ
Questions about on-premise sizing
How accurate is the configurator?
It gives a sizing estimate. Demand comes from measured TextCortex enterprise usage. Server capacity is measured in public benchmarks where marked and estimated otherwise. Request a quotation to confirm the sizing for your organisation.
Which models are included?
DeepSeek V4 Flash, DeepSeek V4.1 Flash, GLM-5.3, GLM-5.3-Flash, MiMo-V2.6-Pro, Kimi K3, Gemma 4 31B, Qwen3.8 27B and Nemotron 3 Super. Each model needs a minimum amount of GPU memory, so not every model fits every server tier.
What does the electricity estimate include?
Servers are assumed to draw 60% of their maximum power around the clock (30% for a standby node), plus 40% for cooling. The price is the average business electricity price for the selected country, excluding VAT.
Can I add a standby node for high availability?
Yes. Turn on the standby node in the server panel. It adds one extra node, and the estimate includes its idle power.
How do I get a price quotation?
Select “Get a price quotation” in the configurator. You can email the configuration to our team or book a call.
Plan your private AI deployment.
Talk to our team about hardware, models and the rollout in your organisation.