Set up Locul your way. Choose where your prompts go.
Locul keeps your firm's memory and bots in one desktop app. The model that answers can live in four places, and each guide below walks you through one of them.
Quick answer
Pick the path by where your prompts may go. Frontier: connect Claude Code, ChatGPT or Codex over MCP, quickest to start, prompts go to US providers. In your country: add a key from an EU or in-country provider such as Scaleway, Mistral or Infomaniak. On-premise: run Ollama or vLLM on a server in your office. Local: run a small open model on one laptop for a free, offline trial.
Which setup path fits your firm?
Start with the question your compliance lead will ask first: where may a client's name, a contract clause or a payroll figure travel? Then pick the path that answers it.
Claude Code, ChatGPT and Codex
Plug Locul's memory into the AI apps your team already uses, over MCP. Takes a few minutes and needs no hardware.
Prompts go to US providers Open the guide → 02 / IN YOUR COUNTRYEU and in-country providers
Use open and European models hosted by Scaleway, Mistral, OVHcloud, IONOS, Infomaniak or Opper with your own API key.
Prompts stay in your country, or at least in the EU See the providers ↓ 03 / ON-PREMISEYour own server with Ollama or vLLM
One GPU server in your office or rack serves the whole team. Includes sizing by team size and EU GPU rental for a pilot.
Prompts stay on your network Open the guide → 04 / LOCAL LAPTOPRun an LLM locally with Ollama
Pull an open model from Hugging Face onto one Mac or PC and point Locul at it. Free, offline, ideal for a first trial.
Prompts never leave the laptop Open the guide →Which providers keep prompts in your country, or in the EU?
Each of these runs models in its own European data centres. You open an account, create an API key and paste it into Locul. No US big-tech cloud sits in the middle.
Scaleway
Generative APIs from a French cloud, hosted in Paris.
Scaleway guide → FRMistral AI
The French model maker's own API for Mistral models.
Mistral guide → FROVHcloud
AI Endpoints serving open models on French cloud.
OVHcloud guide → DEIONOS
AI Model Hub running in German data centres.
IONOS guide → CHInfomaniak
AI Tools hosted in Swiss data centres.
Infomaniak guide → SEOpper
Swedish platform that routes to models hosted in the EU.
Opper guide →Also in the app: Nscale for UK data centres and STACKIT for German data centres. Pick your country in Locul and it lists the providers that fit.
How do the four paths compare?
The short version: the further the model sits from your desk, the less you manage and the more you trust someone else's contract.
| Path | Where prompts go | Who runs the model | You need | Good for |
|---|---|---|---|---|
| Frontier | Anthropic or OpenAI, in the US | The AI company | A Claude or ChatGPT plan | Work that never touches client data |
| In your country | An EU or in-country data centre | A European provider, under your contract | An account and an API key | Daily client work without hardware |
| On-premise | A server on your own network | You, or a Locul partner for you | A GPU server, owned or rented | Teams with strict data rules |
| Local laptop | Nowhere, it stays on the device | You, on one machine | A laptop with 16 GB of memory or more | A free trial, or one person working offline |
Common questions
Which Locul setup path is the most private?
Running a model on one laptop or on a server in your own office is the most private, because prompts never leave your machine or your network. An EU or in-country provider keeps prompts inside your jurisdiction under a contract with that provider. Frontier models such as Claude and ChatGPT send prompts to US companies.
Can I mix setup paths?
Yes. Many teams test on one laptop first, then move to an in-country provider or an office server for daily work, and keep a frontier model for tasks that do not touch client data. Each bot in Locul can use a different model.
Does my Locul memory go to the model provider?
Your memory stays in Locul. Only the prompt and the pieces of memory needed to answer it are sent to whichever model you chose, so on the local and on-premise paths nothing leaves your machine or network.
Do I need a GPU to start?
No. A recent laptop with 16 GB of memory runs small open models well enough for a trial, and the frontier and in-country paths need no hardware at all. A GPU server only matters when a whole team shares one on-premise model.
Try it on one laptop today, decide the rest later.
Download Locul, run a small open model next to it, and see how your own documents answer back. When the team is ready, a partner in your country can set up the server.