IONOS AI Model Hub: open models from German data centres, step by step
The IONOS AI Model Hub serves open models from Berlin with prompts that are never logged. Here is how to get the token, test it and connect it to Locul.
Quick answer
In the IONOS Data Center Designer, open Management > Token Manager, set a time to live and press Generate Token. Copy it, as it will not be shown again. The user needs the group permission Access and manage AI Model Hub. The base URL is https://openai.inference.de-txl.ionos.com/v1 and the model is Qwen/Qwen3.5-397B-A17B. In Locul, open Settings > Model > Open models, pick IONOS, paste the token and press Test. IONOS processes prompts in Germany and never logs them.
Who is the IONOS AI Model Hub a good fit for?
The AI Model Hub suits German firms, such as tax advisers and law firms, that want prompts processed only in Germany by a German provider.
IONOS is a German hosting and cloud company. The AI Model Hub is part of its public cloud: open models behind an OpenAI-compatible API, run in its Berlin data centre (de-txl in the address). Its data handling page is one of the clearest of any provider.
"The content of your prompts and of the outputs generated for you is never logged."
IONOS docs, AI Model Hub data handling
"No third-party sub-processors are engaged for AI Model Hub."
IONOS docs, AI Model Hub data handling
The same page says your data is not used to train or improve any model, that prompts are processed in German data centres and nowhere else, and that IONOS records only metadata: the time, the model and the number of tokens.
| Provider | Country | Where prompts run | Prompts kept? | Used for training? | Pick in Locul |
|---|---|---|---|---|---|
| Scaleway | France | Paris (fr-par) | Not kept by default | No | GLM-5.3 |
| Mistral AI | France | EU endpoint, api.eu.mistral.ai | Zero retention on request | Free plan may; opt out or use a paid plan | Mistral Medium 3.5 |
| OVHcloud | France | Gravelines | Not stored | Not stated; data not stored or shared | Qwen3.5 397B |
| IONOS (this guide) | Germany | Berlin | Never logged | No | Qwen3.5 397B |
| Infomaniak | Switzerland | Swiss data centres | Not recorded | No | Qwen3.5 397B |
| Opper | Sweden | Stockholm gateway, EU model routes | Usage metadata only | No | DeepSeek V4 Pro (TensorX, EU) |
How do I get an IONOS AI Model Hub token?
Unlock the IONOS public cloud, give your user AI Model Hub access, then generate a token in the Token Manager.
Unlock the full account
A new IONOS Cloud contract starts limited. Add payment details and ask IONOS to unlock the full version, which can take up to 24 hours.
Grant AI Model Hub access
In the Data Center Designer (dcd.ionos.com), open Management > Users & Groups. Put the user in a group with the permission Access and manage AI Model Hub. IONOS suggests a dedicated user for API access.
Generate the token
Open Management > Token Manager, choose a Time to Live (TTL) and press Generate Token.
Copy or download it
Download the token or copy it now. IONOS will not show it again in any dialog.
Watch the time to live. When the TTL runs out, Locul's check turns to "did not accept your access key". Set a TTL that matches your key rotation policy and put a reminder in the calendar.
What is the IONOS AI Model Hub base URL?
The OpenAI-compatible base URL is https://openai.inference.de-txl.ionos.com/v1.
Test the token before you open Locul:
# paste your key into the variable first, never into a shared file
$ export IONOS_API_TOKEN="paste-your-key-here"
$ curl https://openai.inference.de-txl.ionos.com/v1/chat/completions \
-H "Authorization: Bearer $IONOS_API_TOKEN" \
-H 'Content-Type: application/json' \
-d '{"model":"Qwen/Qwen3.5-397B-A17B","messages":[{"role":"user","content":"Reply with one word: ready"}]}'A working token returns JSON with a choices list:
{"object":"chat.completion","model":"Qwen/Qwen3.5-397B-A17B",
"choices":[{"index":0,"message":{"role":"assistant","content":"Ready"},"finish_reason":"stop"}],
"usage":{...}}Which IONOS model should I pick in Locul?
Pick Qwen3.5 397B: IONOS lists it with a 262k-token context window and tool calling, which Locul's bots rely on.
| Model ID | Context | Use it for |
|---|---|---|
Qwen/Qwen3.5-397B-A17B | 262k tokens | Recommended. Long files, receipts, bots with tools |
openai/gpt-oss-120b | OpenAI's open-weight model, a second option |
IONOS retired Mistral Nemo on 22 October 2026, so Locul no longer lists it.
How do I add IONOS to Locul?
Open Settings > Model > Open models, pick IONOS under In your region, paste the token and press Test.
Open Settings, then Model
In the Locul desktop app, open Settings and choose Model.
Choose Open models
Pick Open models (marked Recommended), then find IONOS under In your country. Each row shows where the provider runs and whether it trains on your data.
Pick the model
Under Model, choose Qwen3.5 397B. Locul tags it Recommended.
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 IONOS if you have not made one yet.
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.
Where the key lives. Locul saves it in your Mac's Keychain. It never shows the key again and sends it nowhere except IONOS.
Why does the IONOS connection check fail?
Usually because the token's time to live has run out, or the user is not in a group with AI Model Hub access.
| Code | Locul says | What to do |
|---|---|---|
| 401 | "IONOS did not accept your access key." | Check the token has not expired, and that its user is in a group with Access and manage AI Model Hub. Generate a new token in Token Manager if needed. |
| 404 | "IONOS does not offer Qwen/Qwen3.5-397B-A17B for your account." | Check the model is still on the AI Model Hub model list; IONOS retires older models with notice. |
| 402 | "IONOS says your account has no credit left." | The contract may still be limited. Add payment details and ask IONOS to unlock the full version. |
| 429 | "IONOS is asking us to slow down." | Wait a minute and press Test again. For steady heavy use, ask IONOS support about limits. |
| No answer | "Could not reach IONOS." | Your network blocks openai.inference.de-txl.ionos.com. Ask IT to allow HTTPS to that address. |
| Format | "That does not look like a whole access key." | The key is one long line with no spaces. Copy it again in one go, without a line break at the end. |
What does a real request on IONOS look like?
A messy, everyday job with client names in it, the kind of work this setup is for.
Here is a request a tax adviser in Cologne might send to a Locul bot running on IONOS:
The receipts and the client's name are processed in IONOS data centres in Germany, and IONOS says the prompt content is never logged.
How fast and how good is IONOS in Locul?
We are timing real office tasks on IONOS with the recommended model, and the results will be published here.
Results coming
Response times, tool use and answer quality on real office tasks, measured on IONOS. Coming soon.
What other in-country providers work with Locul?
Five more EU and Swiss providers, each with its own guide.
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 AI guide →FROVHcloud
AI Endpoints serving open models on French cloud.
OVHcloud guide →CHInfomaniak
AI Tools hosted in Swiss data centres.
Infomaniak guide →SEOpper
Swedish platform that routes to models hosted in the EU.
Opper guide →Frequently asked questions
What is the IONOS AI Model Hub?
It is IONOS's managed AI service: open models such as Qwen3.5 behind an OpenAI-compatible API, run in IONOS data centres in Germany. You pay per use; see IONOS's AI Model Hub page for pricing.
Where is the IONOS AI Model Hub hosted?
In IONOS data centres in Germany, in Berlin. IONOS says prompts and outputs are processed there and nowhere else.
Does IONOS log my prompts?
No. IONOS says the content of prompts and outputs is never logged or written to storage. It records only metadata: the time, the model and the token count.
Does IONOS use my data to train models?
No. IONOS says data sent to the AI Model Hub is not used to train, fine-tune or improve any model, and no third-party sub-processors are involved.
Why does my IONOS token stop working?
Tokens have a time to live set when you generate them. Once it runs out, make a new token in the Token Manager and paste it into Locul.
Sources
Every command and quote on this page comes from these primary sources, checked on 24 October 2026.
Junaid builds Locul and sets up private AI for small regulated firms. He tests every command on this page on his own machines before it ships.