Geeps is a BYOK chat app for iPhone, iPad, and Mac. It can connect to custom APIs that use the OpenAI Chat Completions format, which makes it a simple way to use LLMBase models in a native Apple-device chat client.
This setup uses the direct LLMBase Inference API. It gives Geeps its own inference key and uses prepaid API credits. If you need an EU-hosted model, choose one that is currently labelled EU-hosted in the LLMBase model catalog. That selection is important: model processing information depends on the model you choose, not merely on the app connection.
Before you start
You need:
- Geeps on an iPhone, iPad, or Mac.
- An LLMBase inference API key beginning with
llmbase_.... - A positive prepaid inference balance.
- A model from the current LLMBase catalog.
Do not use a llmbase_chat_... agent key in Geeps. Geeps is configured here as a direct OpenAI-compatible API client, so it needs an llmbase_... inference key.
The LLMBase values to enter in Geeps
Geeps builds the final Chat Completions path from the base host. Use these values exactly:
| Geeps field | Value |
|---|---|
| Provider name | LLMBase |
| API key | Your llmbase_... inference key |
| Base URL | api.llmbase.ai |
| Full custom endpoint | Leave empty |
| Default model | An exact ID from the LLMBase model catalog |
With that base URL, Geeps forms the expected request URL:
https://api.llmbase.ai/v1/chat/completionsUse either Base URL or Full custom endpoint. Entering both can create an invalid endpoint preview. For the standard LLMBase setup, the Base URL field is all you need.
Add LLMBase as a custom provider in Geeps
- Open Geeps and go to Settings → API Settings.
- Choose Add Custom Provider.
- Give the provider the name
LLMBase. - Paste your LLMBase inference key.
- Enter
api.llmbase.aias the Base URL. - Leave Full custom endpoint blank.
- Enable the provider and save the settings.
- Use the endpoint preview to check that it ends in
/v1/chat/completions.
Geeps can load models from an API that exposes /v1/models. After saving the provider, use the model reload action for LLMBase. If a model does not appear, enter its exact public LLMBase model ID manually, then test the key before starting a real conversation.
Choose the right model for the task
The best default model is not always the largest one. Start with the work you expect to do in Geeps:
- Everyday writing and answers: choose a capable text model with an appropriate context window.
- Code or complex analysis: use a model whose published capabilities suit the task and test it with a representative prompt.
- Images or document context: verify that the model supports the relevant input type before relying on it.
- EU-hosted work: select an LLMBase model explicitly marked EU-hosted in the catalog.
For an EU-hosted workflow, avoid relying on a previously selected default forever. Review the chosen model before work that has a strict location or policy requirement, and use the current published LLMBase information as the basis for an internal approval decision.
Make Geeps useful as a personal AI workspace
Once LLMBase is connected, configure one simple workflow before changing lots of settings. For example:
- Set a model as the LLMBase default provider.
- Save a reusable prompt for a recurring job, such as meeting notes, code review, or a research brief.
- Start a new chat for each unrelated task instead of changing the model mid-conversation.
- Use a short system instruction that tells the model what a useful answer looks like.
Try this first prompt:
You are my concise research assistant. Separate verified facts from suggestions, identify missing context, and give me the three most useful next questions to ask.
That produces a baseline you can evaluate before adjusting temperature, output length, or other model settings. Only enable a setting when the chosen model supports it.
Protect your API key on a personal device
An API key is a credential, not a shareable account password. It can spend from its associated inference balance. Use a dedicated LLMBase key for Geeps rather than sharing a key with an application or a team deployment.
Good habits include:
- Keep device access protected with a passcode and your normal device security controls.
- Do not paste the key into screenshots, shared notes, or chat messages.
- Set a spend cap for the key where appropriate.
- Revoke the key in LLMBase immediately if the device, backup, or key may have been exposed.
- Create separate keys for separate devices or use cases when you need cleaner revocation and usage review.
Troubleshooting Geeps and LLMBase
| Problem | Fix |
|---|---|
| The API test fails | Check that the key starts with llmbase_..., the account has inference credit, and there are no spaces before or after the key. |
| The endpoint preview is wrong | Enter api.llmbase.ai only in Base URL and leave the full endpoint empty. |
| Models do not load | Reload the LLMBase model list. If needed, copy an exact public model ID from the catalog and enter it manually. |
| A feature does not work | Confirm the selected model advertises the required capability, such as image input or tools. |
| You need EU hosting | Choose a model currently marked EU-hosted in the catalog; a generic custom-provider setup is not itself a location setting. |
FAQ: Geeps with LLMBase
Does Geeps work with LLMBase?
Yes. Geeps supports custom providers that use OpenAI-compatible Chat Completions, and LLMBase provides that API format.
Which endpoint should I use in Geeps?
Enter api.llmbase.ai in Geeps’ Base URL field. Geeps constructs https://api.llmbase.ai/v1/chat/completions from it.
Can I use a Chat/Agent subscription key in Geeps?
No. Use an llmbase_... inference API key for Geeps. Chat-agent keys have a different endpoint and are intended for supported agent integrations.
For screenshots and the underlying setup reference, see the LLMBase Geeps integration documentation and the API authentication guide.
