Open WebUI is a flexible interface for people who want to choose where their chat models come from. LLMBase works with it through the standard OpenAI-compatible Chat Completions API: add one connection, choose a model, and use Open WebUI as normal.
This guide is for the direct LLMBase Inference API. It is a good fit for a personal or team Open WebUI deployment that needs its own API key and prepaid inference balance. If EU hosting is a requirement, select a model marked EU-hosted in the LLMBase model catalog before putting the connection into use. Model processing location is model-specific, so do not treat a general API connection as a location guarantee for every model.
What you need before connecting Open WebUI
Prepare these three things:
- An LLMBase inference API key beginning with
llmbase_.... - The LLMBase API base URL:
https://api.llmbase.ai/v1. - A model ID from the current LLMBase model catalog or API model list.
Do not use an llmbase_chat_... chat-agent key here. That key is only for the separate agent endpoint used by tools such as OpenClaw and Hermes. Open WebUI’s normal OpenAI-compatible connection should use an inference key and the /v1 base URL.
Create the key in the LLMBase dashboard and keep it in a secret store or the configuration mechanism appropriate for your Open WebUI deployment. Do not paste a production key into a public Docker Compose file, a checked-in .env file, a browser setting that other people can read, or a screenshot.
Add LLMBase as an OpenAI-compatible connection
Open WebUI’s current connection flow is in Admin Settings. The exact labels can change between releases, but the path is typically Admin Settings → Connections → OpenAI.
- Open Manage or the connection settings for OpenAI.
- Select Add Connection.
- Choose an external OpenAI-compatible connection.
- Enter the following values:
| Open WebUI field | Value |
|---|---|
| Connection name | LLMBase |
| URL | https://api.llmbase.ai/v1 |
| API key | Your llmbase_... inference key |
| Model filter | Leave empty initially so Open WebUI can discover available models |
- Save the connection.
- Open a new chat and choose an LLMBase model from the model picker.
Open WebUI normally discovers models through the OpenAI-compatible /models endpoint. If your deployment uses a model allowlist, add only the exact public LLMBase IDs you have evaluated. Avoid copying a model name from an old article or an unrelated provider’s list.
Select an EU-hosted model deliberately
An Open WebUI connection and a model choice are separate decisions. The connection tells Open WebUI where to send a request; the selected model determines the capabilities and model-specific processing information for that request.
For work that requires EU hosting:
- Open the LLMBase model catalog.
- Filter or identify models labelled EU-hosted.
- Confirm that the model supports the task you need, such as text, tools, long context, or image input.
- Add that exact model ID to Open WebUI’s allowlist if you want to prevent accidental selection of another model.
- Review the current published privacy and contractual documentation with the people responsible for your policy requirements.
This is more reliable than relying on a vague provider label. It gives the team a concrete model choice to review, document, and revisit when requirements change.
A sensible first test
Start with a harmless prompt before connecting real project material:
Reply with the current model ID you received and write a three-sentence summary of the difference between a project brief and a project plan.
Then check the selected model name, the response, and the API usage in the LLMBase dashboard. If the model supports the capability you need, test one representative workflow next: a short writing task, a code review, a source-backed research question, or an approved file analysis.
Do not enable every Open WebUI feature just because a connection works. Features such as tool use, image input, structured output, and long-context tasks depend on the selected model. Use the model catalog metadata to choose a compatible model before sending a production request.
Keep the Open WebUI connection operationally clean
For a team deployment, treat the provider connection as an administrative integration:
- Use a dedicated key for Open WebUI so it can be revoked without affecting another application.
- Set a sensible spend cap on the inference key where appropriate.
- Restrict who can edit provider connections and model allowlists.
- Keep a short record of the approved model IDs and the work they are intended for.
- Rotate or revoke a key immediately if it may have been exposed.
Open WebUI controls who can use its interface and which connections it exposes. LLMBase controls the API key and model request. Keeping those responsibilities distinct makes troubleshooting and access review much easier.
Common Open WebUI and LLMBase errors
| Problem | What to check |
|---|---|
401 or invalid API key | Confirm the key starts with llmbase_..., has no extra spaces, and is paired with https://api.llmbase.ai/v1. |
| No models appear | Save the connection, then refresh the model picker. If an allowlist is enabled, add exact LLMBase model IDs. |
A request returns 400 | The selected model may not support a requested field or feature. Check its current capabilities before enabling that feature in Open WebUI. |
| A request cannot be made from your server | Confirm the Open WebUI host can make outbound HTTPS requests to api.llmbase.ai. |
| An EU-hosted model is required | Choose a model explicitly marked EU-hosted in the catalog; do not assume every model has the same processing location. |
FAQ: using LLMBase with Open WebUI
Is Open WebUI compatible with LLMBase?
Yes. Open WebUI supports OpenAI-compatible API connections, and LLMBase provides an OpenAI-compatible Chat Completions API at https://api.llmbase.ai/v1.
Which LLMBase key should I use in Open WebUI?
Use an llmbase_... inference API key. It uses prepaid inference credits. The llmbase_chat_... key belongs to LLMBase’s separate agent endpoint and is not for this connection.
Can I force Open WebUI to use only EU-hosted models?
Yes, use Open WebUI’s model allowlist with exact model IDs that are currently marked EU-hosted in the LLMBase catalog. Recheck the catalog and your approved documentation when you update the allowlist.
For endpoint details and model discovery, see the LLMBase Quickstart, OpenAI compatibility reference, and model discovery guide.
