Elephas

AI Provider Configuration

Step-by-step setup for OpenAI, Claude, Gemini, Ollama, and other AI providers in Elephas. Switch between Offline, Cloud, and Bring Your Own Key modes.

Choosing your AI provider

Elephas supports three modes for AI processing. You can switch between them at any time.

Offline AI

Runs a language model locally on your Mac. Requires Apple Silicon. Best for maximum privacy. No internet connection needed after the model is downloaded.

Cloud AI (Elephas proxy)

Uses OpenAI or Claude through the Elephas reverse proxy. You do not need your own API key. Your data is not used for training.

Bring Your Own Key

Connect your own API key from OpenAI, Anthropic (Claude), or Google (Gemini). Requests go directly from your Mac to the provider.

How to change providers

  1. Open Elephas Settings
  2. Go to the AI Provider tab
  3. Select your preferred mode
  4. If using Bring Your Own Key, paste your API key
  5. Click Save

Setting up OpenAI

  1. Go to platform.openai.com/api-keys
  2. Create a new API key
  3. Copy the key and paste it in Elephas Settings under AI Provider > OpenAI
  4. Select your preferred model (GPT-4o recommended for best quality)

Your OpenAI API key is separate from a ChatGPT subscription — API usage is pay-as-you-go, billed by token usage. To use GPT-4 / GPT-5 models, your OpenAI account needs a payment of at least $1 (a card on file). If the models don't appear in Preferences, add billing and restart Elephas. You choose which model each feature uses under Model Settings and in Super Chat's chat settings. Keep your API key private.

Setting up Claude (Anthropic)

  1. Go to console.anthropic.com
  2. Create an API key
  3. Paste it in Elephas Settings under AI Provider > Claude
  4. Select your preferred model (Claude Sonnet recommended for everyday use)

Setting up Gemini (Google)

  1. Go to aistudio.google.com
  2. Create an API key
  3. Paste it in Elephas Settings under AI Provider > Gemini

Setting up Ollama (local)

Ollama lets you run open-source models locally. Install Ollama from ollama.com, then pull a model (e.g., ollama pull llama3). Set it in Elephas under Settings > Offline AI > the gear icon next to Ollama; running models are detected automatically. Keep Ollama running in the background.

To index files into Super Brain with Ollama, also pull an embedding model: ollama pull nomic-embed-text. Good chat models include Llama 3.2, Mistral, and DeepSeek (e.g. ollama run deepseek-r1:7b). On Apple Silicon (M1+) you can instead use Elephas's one-click built-in offline models when creating a Brain — no Ollama needed. See Offline AI.

Managing offline models

In Preferences > Offline you manage two groups, Chat Models and Indexing Models:

  • Download: click Download, wait, then Start to load it locally (shows Running)
  • Only one offline model runs at a time — starting a new one auto-stops the current one
  • Stop: the red Stop button returns it to Ready to start
  • Delete: the Trash icon (with confirm); you can re-download later

Verifying an OpenAI key that isn't working

If your OpenAI key is rejected, the usual causes are a mistyped or deleted key, or a region-based access restriction on your OpenAI account. You can test the key directly from Terminal:

curl -u :YOUR_KEY https://api.openai.com/v1/models

Replace YOUR_KEY with your sk- key. A JSON list of models means the key works; an error response means the key or account is the problem. (See also You Exceeded Your Current Quota for billing errors.)

Deleting a downloaded model (including Whisper)

Offline chat, indexing, and Whisper (audio) models are stored in the app's Library folder. Offline chat/indexing models can be removed from Preferences > Offline (Trash icon). To remove a downloaded Whisper model manually, open Finder, press Cmd + Shift + G, go to ~/Library/, and delete the model file. You can re-download it later.

The built-in Elephas engine and credits

If you don't bring your own key, the built-in Elephas engine routes to leading cloud models (from OpenAI, Anthropic, Google, and others) using monthly credits that reset each cycle — higher plans include more credits (see Billing, Plans & Credits). Claude Haiku is a good default for most writing. The exact model list is shown in Model Settings, where you pick a model per feature.

More BYOK providers

Elephas supports many providers. For most, you create a key on the provider's site, then open Preferences > AI Providers, select the provider, click its gear icon, paste the key, and Verify. Then pick its models in Model Settings / Super Chat.

  • Groq (fastest inference): key at console.groq.com/keys. Hosts Llama 3, Mixtral, Gemma. Also the easiest way to run Llama 3 in Elephas.
  • Perplexity: add payment, then key at perplexity.ai/settings/api.
  • Together AI: key at api.together.xyz/settings/api-keys.
  • OpenRouter: key at openrouter.ai/settings/keys (endpoint https://openrouter.ai/api).
  • DeepSeek: key from the DeepSeek platform (API Keys > Create); or run it offline via Ollama.
  • Grok (xAI): key at console.x.ai (API keys > Create API Key).
  • Fireworks AI: configured under the Custom AI provider (not its own section). Endpoint https://api.fireworks.ai/inference plus your key; then pick it from “custom AI” in Model Settings.

Voyage AI (for indexing)

Voyage provides embedding models used to index your Super Brain. Create a key in your Voyage account (API key > Create new secret key), then in Preferences > AI Providers click the gear next to Voyage AI, enter the key, and Save. Select Voyage as the Indexing Model in Model Settings (default for all Brains, or per-Brain in its settings).

Azure OpenAI

Create the Azure resource first:

  1. In the Azure Portal, Create a Resource > search “Azure OpenAI” > Create
  2. Choose Subscription / Resource Group, set a Resource Name (used in the Endpoint URL), and pick a Region that supports your model
  3. Review + Create, then open Keys and Endpoint and copy the Endpoint (Resource URL) and an API Key
  4. Under Deployments, create a deployment of your model (e.g. gpt-4) — Elephas can only use deployed models

Connect it in Elephas: Preferences > AI Providers > Azure OpenAI > gear icon > enter the Resource URL and API Key (Azure needs both) > Verify. Elephas loads your deployed models; choose which to use and Save. If no models appear, recheck the endpoint/key and confirm the deployment is active, then re-verify.

Related articles

Privacy & Data ProtectionOffline AI ModeRun language models locally on your Mac with no internet connection. Complete privacy with Apple Silicon.Privacy & Data ProtectionCloud AI and Your DataUnderstand exactly what is sent to OpenAI, Claude, and Gemini when you use cloud AI mode in Elephas.SettingsKeyboard ShortcutsCustomize Super Command shortcuts and other key bindings in Elephas. View default shortcuts and troubleshoot conflicts.