Your staff already use ChatGPT, mostly because it is quick. Locul is just as quick in the browser, but it runs top open models on servers for your firm alone and answers from your own files, with a source on each answer. Nothing is sent to OpenAI or anyone else.
ChatGPT is not private by default. On the free and Plus plans, chats can be used to train models unless you switch it off. Business and Enterprise plans do not train on your data, but every prompt is still processed by OpenAI.
A private ChatGPT for a team is a ChatGPT-style chat running an open model on servers you control or rent for yourself alone. Nothing is processed by an AI company, and it can answer from your own files.
The detail changes, so check OpenAI's current terms before you rely on any of it. The shape has stayed the same for a while.
| ChatGPT Free and Plus | ChatGPT Business and Enterprise | Azure OpenAI | Locul | |
|---|---|---|---|---|
| Used to train models | By default, unless switched off | No, by default | No | Never |
| Who processes the prompt | OpenAI | OpenAI | Microsoft | Servers for your firm alone |
| Where | Mainly the US | Regional storage on some plans | The Azure region you pick | UK, EU or your building |
| Shared with other customers | Yes | Yes | Yes, unless you pay for dedicated capacity | No |
| Answers from your files, with permissions | No | Through connectors | You build it | Yes, with a source on every answer |
| Runs with no internet | No | No | No | Yes, on your own servers |
Most questions at work are not "write me a poem". They are "which of these four files is the right one, and fast". A private ChatGPT that can read your files, and knows who is asking, answers that.
kurze Frage, which version of the Halvorsen framework agreement is the live one?? Lena says v7 but the folder has v7_final, v7_final_FINAL and 'v8 DRAFT do not send'. Customer wants the liability cap clause by 3pm their time, so 2pm ours.
The live one is v7_final_FINAL, signed 2 September 2026. v7_final was replaced the same day after Halvorsen's legal team changed the payment terms in clause 9. 'v8 DRAFT' is a proposal from Lena from October and has not been agreed.
The liability cap is clause 14.2: 150% of the fees paid in the previous 12 months, excluding data protection claims, which are uncapped. Do you want a short email to the customer quoting it in English and German?
An example. The company, customer and people are invented.
For the drafting, summarising and document questions that make up most office work, the gap is small. The benchmark shows how small.
Banning ChatGPT does not work. People use it on their phones instead. What works is giving them something just as easy, then a short policy.
Same sign-in as everything else, nothing to install.
Client and personal data go in the private tool. Public AI is fine for things you would post on your website.
The inbox, the policy questions and the file nobody can find. Word of mouth does the rest.
See what people actually ask, then add bots for the questions that keep coming back.
We test open models on real office work: messy emails, spreadsheets with broken formulas, forwarded chains with the ask buried in paragraph four. See how close they get to the big cloud models.
Not from OpenAI in the strict sense: every ChatGPT plan processes prompts on OpenAI's infrastructure. Business and Enterprise plans do not train on your data by default and offer some regional storage, which suits many firms. A private ChatGPT-style tool runs an open model on servers you control or rent for yourself alone, so prompts are never processed by a third party.
Paid plans give you more control. ChatGPT Business and Enterprise exclude your data from training by default and add admin controls. The prompts are still processed by OpenAI, a US company, so you rely on the contract rather than on where the data physically goes.
ChatGPT has Temporary Chat, which is not saved to history or used for training, and a setting to stop your chats being used to improve models. Both reduce what is kept. Neither changes where the prompt is processed.
You need a chat app, an open model, a search index over your documents that respects who may see what, and servers to run it on. You can build that yourself (see our self-hosted LLM guide) or have Locul set it up on your servers or a dedicated one in your region.
For most office work, close. The best open models handle drafting, summarising, translation and questions over your own documents well. We publish the gap on real office tasks in the Office AI Benchmark, so you can judge for yourself.
One call. We show Locul answering from files like yours, on servers like the ones it would run on for you.