Private ChatGPT for your team, on a server you control

A private ChatGPT is not a different model. It is the same kind of chat workspace your team already uses, running on an instance you control, talking to a model you chose, with the history stored somewhere that belongs to you. This is what that looks like when we run it for you.

What “private ChatGPT” actually means

Three pieces make up any chat assistant: the interface people type into, the model that writes the answers, and the storage that keeps accounts, conversations and uploaded files. A consumer subscription bundles all three with one vendor. A private setup separates them.

The interface is open-source software — Open WebUI or LibreChat — on a dedicated instance with your team’s accounts. The model is reached through an API key you own, or runs on a server of your own. The storage is that instance’s database, backed up off the machine every night. Nobody else’s workspace ever holds your prompts.

Why teams move off individual subscriptions

It rarely starts as a policy decision. It starts with a handful of people on personal plans, a customer contract pasted into a consumer chatbot, and nobody able to say afterwards which account saw it. A shared, self-hosted workspace answers all of that at once: one login per person, one place where the model keys live, and a record your company owns.

The pricing shape changes too. Per-seat plans charge for every person whether they send one message a month or five hundred. A private instance is a flat monthly price, and model usage is billed by the provider at list price for what is actually used. For teams where most people are light users, that difference is usually the whole argument.

What you get when we host it

A dedicated instance, not a shared multi-tenant account: your interface, your database, your files, isolated from other customers.

Your domain or ours, with TLS, so the team opens something like chat.yourcompany.com rather than a vendor URL.

User management built into the software: invite people, set roles, decide who can add models or upload documents. Both Open WebUI and LibreChat support OpenID Connect if you want sign-in through an identity provider you already run.

The operational part: a version pinned to a reviewed release and updated deliberately, nightly off-box backups, and a human to write to when something looks wrong. That last part is the reason to pay for hosting rather than run a container yourself.

Open WebUI or LibreChat

Open WebUI is where most teams should start. It is the most complete interface in the category, handles document upload and retrieval well, works equally with commercial APIs and local models, and runs comfortably on a small plan. Its licence is BSD-3 with a branding clause — the Open WebUI branding stays visible, which matters only if you intended to white-label it.

LibreChat is the better fit when the team switches between many providers in the same conversation view, wants agents and MCP tools inside the chat, or needs an MIT-licensed codebase. It needs a little more memory, so plans start at 4 GB.

If you are unsure, tell us how the team works and we will set up the one that fits. Moving later is possible but means exporting and re-importing history, so it is worth deciding once.

Choosing the model behind it

A commercial API key — OpenAI, Anthropic, Google, Mistral — gives the strongest answers and is what most teams use day to day. Business API traffic is handled under the provider’s API terms rather than consumer-app terms; read your provider’s data policy, but for most this is a meaningful step up from staff pasting documents into a personal account.

Several providers at once is where LiteLLM comes in: one gateway in front of every provider, with budgets per team and a spend log you can query. The chat interface talks to the gateway; finance talks to the log.

Nothing leaving your servers means a local model. An Ollama plan runs 7-8B models on CPU on the same private network. Expect it to be slower and less capable than a frontier model — good for summaries, drafts and sensitive material, not for hard reasoning. Many teams mix the two: local for the confidential work, a commercial key for everything else.

What it costs

A team of ten on Open WebUI typically lands on the 4 GB plan at $9.99 a month. Model usage is paid to the provider directly: for light, conversational use across a small team that is often a few dollars to a few tens of dollars a month, and the cost calculator will give you your own number from real request volumes.

Add LiteLLM from $5.99 if you want budgets and a spend log, and an Ollama plan from $16.99 if some of the work should never reach an external API. We do not proxy or mark up model usage; the hosting price is the whole of what you pay us.

How the setup works

Pick the interface and plan, and tell us the domain you want it on. We provision the instance, issue the certificate, create the first admin account and hand it over. You add your model key in the admin panel — it never has to pass through us — invite the team, and they are working the same day.

From then on updates, backups and monitoring are ours. If you outgrow the plan, it is resized in place; the history and accounts stay where they are.

Related reading

Open WebUI

A ChatGPT-style workspace for your team, pointed at any model you like.

from $5.99/moDetails →

LibreChat

Multi-model chat with agents, plugins and conversation forking.

from $11.99/moDetails →

LiteLLM

One OpenAI-compatible endpoint in front of every model provider you use.

from $5.99/moDetails →

Ollama

Your private LLM server — pull a model, get an API, keep every token on your own box.

from $16.99/moDetails →