You are not signed inLog inCreate account

shop.antonnovikov.com / private-llm

Private LLM on your own server

Some data cannot be sent to an outside API: contracts, customer records, source code. I size the hardware, then deploy an open-weights language model on your own server or a rented GPU, with an OpenAI-compatible API, access control, monitoring and a written update procedure. The data stays with you, and you pay for the hardware, not for every request.

from $2,850 + $1,190/motypically 10–14 daysAI models are paid separately, at the provider’s price

What is included

  • GPU sizing report: which model fits which hardware, measured speed in tokens per second, and a cost comparison of buying versus renting
  • Deployment as code in your repository (Docker Compose or Helm), so the whole setup can be rebuilt with one command
  • An OpenAI-compatible API behind TLS with authentication: tools and SDKs that speak the OpenAI API work after changing the base URL
  • Logs, health checks and a basic dashboard: GPU load, memory, latency and tokens per second
  • A written update and rollback procedure for the model, the serving software and the GPU drivers
  • Handover: source code and a short operations guide

Calculator

“+$” adds to the price, “+N%” scales the total, “/mo” is paid after launch.

Model size

The GPU or server is bought or rented separately. The sizing report checks this choice against your real load before anything is deployed.

How the model runs

Ollama is the simplest and fits a small team. vLLM is built for many concurrent users on the same GPU; on Kubernetes it also scales across machines.

What users get

Access control

Operations

Subscription

What a subscription includes, and the terms

A subscription makes the build cheaper: I keep the service running, watch its errors and AI spend and improve it every month, so the build costs less. Without it you get the code and run it yourself at the full price.

Timing

Request

To send a request, log in or create an account: you will see its status and can write to me in one thread. Your chosen options are kept.

Create account Log in

Other services

  1. AI assistant for support and salesfrom $920 A chat assistant that answers customers from your knowledge, on your site or in messengers. 6–9 days · +$790/mo
  2. Knowledge base search and Q&A (RAG)from $1,840 Ask questions in plain language, get answers with sources from your documents. 8–12 days · +$1,190/mo
  3. AI workflow automationfrom $710 Routine work done by n8n or Python workflows with LLM steps: sort, summarise, draft, route. 4–6 days · +$490/mo
  4. AI Telegram botfrom $1,050 A Telegram bot that chats, reads files and photos, remembers context and does things for users. 7–10 days · +$520/mo

Calculator for all services