NirPrivate · Closed-loop AI In active development · We train Nirmind hereYour AI, on your server.
NirPrivate is the private AI machine we built on our DGX systems and our own GPUStack-based orchestration layer. Models run in a closed loop — your data never leaves the server, and there is no token bill.
0 packets
Outbound traffic at start-up and idle — measured
₺0
Token bill · no cloud API
1,597
Automated tests passing
GPU · MLX
NVIDIA GPU and Apple Silicon paths
Our own GPUs
Our DGX machines, with data circling inside a sealed glass box: no way out.
Nirmind learns here
Turkish-first Nirmind models, trained on our own hardware.
Closed loop
Zero bytes leave the server.
A question, a document, audio or an image — your data goes to the model and the answer comes back to you, all inside the same server boundary. Nothing goes out to a third-party API, the open internet or another cloud.
- Inference is fully local
- Your fine-tuning and training data stays with you
- Optional air-gapped (offline) installation
NirPrivate panel
An AI control panel that runs on your own machine.
Try models with chat, embeddings, reranking, image and audio; tune parameters live.
What we're working on right now
Nirmind is training on our own GPUs.
Illustrative view — inspired by the training screen of the NirPrivate control panel.
Training loss
step 48k
Loss
1.050
Tokens / s
182,000
Data
TR · EN · BG · RU
GPU usage
Orchestration queue
● nirmind-tr-v3 · fine-tuning · running
● nirvoice-tts · evaluation · queued
● embed-tr · re-indexing · queued
The power of the cloud. In your own castle.
Closed-loop inference
Models run on your own server; requests never leave it.
Custom model training
Fine-tune on your own data; the weights are yours.
GPU orchestration
A GPUStack-based layer: clusters, worker nodes, deployments, routes.
DGX machines
Training and inference on our own DGX hardware.
Data sovereignty
Auditable logging, in line with KVKK and your organisation's policies.
Nireya integration
Nirmind and NirAIHub send the private route to NirPrivate.
Roadmap
We're still building.
Phase 1 · Done
Platform and infrastructure
DGX machines, a GPUStack-based panel; zero outbound connections, verified.
Phase 2 · In progress
Nirmind training
Turkish-first Nirmind models are training on our own GPUs.
Phase 3 · Up next
Enterprise closed loop
Dedicated, isolated NirPrivate environments for organisations.
Phase 4 · Up next
NirPrivate on your server
A turnkey installation running in your own data centre.