OVHcloud AI Endpoints: open models on French cloud, set up step by step

Who are OVHcloud AI Endpoints a good fit for?

AI Endpoints suit firms that already trust OVHcloud, Europe's largest home-grown cloud, and want open models served from France.

OVHcloud is a French company with its own data centres. AI Endpoints is its serverless model service: no GPU to rent, an OpenAI-compatible API and a catalogue of open models. You use it inside a Public Cloud project, which needs a payment method. Projects still in Discovery mode cannot create keys.

"Our infrastructure, located in Gravelines, France, adheres to European data protection regulations."

OVHcloud docs, AI Endpoints getting started

"Data is not stored or shared during or after model use"

OVHcloud, AI Endpoints

OVHcloud's pages say prompts are not stored or shared. They do not make a separate statement about training, though data that is never stored cannot be reused for it. Your contract terms, including the data processing agreement, are on OVHcloud's contracts page.

How do I create an OVHcloud AI Endpoints API key?

Open your Public Cloud project in the Control Panel, go to AI Endpoints and create a new API key.

1. Open your Public Cloud project

   Sign in to the OVHcloud Control Panel, open Public Cloud and select the project. If you have none, create one and add a payment method.

2. Go to AI Endpoints

   In the left menu, under AI & Machine Learning, choose AI Endpoints.

3. Create the key

   Open the API keys tab and press + Create a new API key. Give it a name such as "Locul, Julien's Mac", an optional description and an expiry date, then press Create.

4. Copy it from the table

   The key appears in the table once. OVHcloud does not keep it for you, so paste it into Locul or your password manager now.

What is the OVHcloud AI Endpoints base URL?

The OpenAI-compatible base URL is https://oai.endpoints.kepler.ai.cloud.ovh.net/v1.

OVHcloud's own examples keep the key in a variable called OVH_AI_ENDPOINTS_ACCESS_TOKEN. Test it before you open Locul:

# paste your key into the variable first, never into a shared file
$ export OVH_AI_ENDPOINTS_ACCESS_TOKEN="paste-your-key-here"
$ curl https://oai.endpoints.kepler.ai.cloud.ovh.net/v1/chat/completions \
    -H "Authorization: Bearer $OVH_AI_ENDPOINTS_ACCESS_TOKEN" \
    -H 'Content-Type: application/json' \
    -d '{"model":"Qwen3.5-397B-A17B","messages":[{"role":"user","content":"Reply with one word: ready"}]}'

A working key returns JSON with a choices list:

{"object":"chat.completion","model":"Qwen3.5-397B-A17B",
 "choices":[{"index":0,"message":{"role":"assistant","content":"Ready"},"finish_reason":"stop"}],
 "usage":{...}}

The Try button on each model in the AI Endpoints catalogue shows the same request with the model's exact name, if you want to copy it from there.

Which OVHcloud model should I pick in Locul?

Pick Qwen3.5 397B: it has a 262k-token context window and function calling, which Locul's bots need for files and memory.

How do I add OVHcloud to Locul?

Open Settings > Model > Open models, pick OVHcloud under In your region, paste the key and press Test.

1. Open Settings, then Model

   In the Locul desktop app, open Settings and choose Model.

2. Choose Open models

   Pick Open models (marked Recommended), then find OVHcloud under In your country. Each row shows where the provider runs and whether it trains on your data.

3. Pick the model

   Under Model, choose Qwen3.5 397B. Locul tags it Recommended.

4. Paste the key

   Under Access key, paste the key into the field and press Save and test. The Get a key link opens the right page at OVHcloud if you have not made one yet.

5. Run the connection check

   Under Connection check, press Test. Ready means chat and tools both work. Ready with limits means the model can chat but could not use tools in the check, so pick another model for bots that read files.

Why does the OVHcloud connection check fail?

Usually because the key was copied from the wrong project, or the project is still in Discovery mode without a payment method.

What does a real request on OVHcloud look like?

A messy, everyday job with client names in it, the kind of work this setup is for.

Here is a request an insurance broker near Lille might send to a Locul bot running on OVHcloud:

The quotes, the claim history and the client's name are processed in Gravelines, France, and OVHcloud says it does not store them.

How fast and how good is OVHcloud in Locul?

We are timing real office tasks on OVHcloud with the recommended model, and the results will be published here.

What other in-country providers work with Locul?

Five more EU and Swiss providers, each with its own guide.

Frequently asked questions

Other setup paths