Providers

A provider is the service that runs the AI model, such as Anthropic, OpenAI, or a model on your own machine. Empryo works with many providers. You need at least one before you can start.

There are three ways to connect. Pick the one that matches what you have.

Quick startLink to this section

You have an API keyLink to this section

Save it with --set-key, using the provider's name:

empryo --set-key anthropic sk-ant-...

Then start Empryo, press Ctrl+L, and pick a model. That is all.

The key is stored in your system keychain. If you prefer, you can also set the provider's usual environment variable (for example ANTHROPIC_API_KEY), or add the key in the app: /keys in the terminal UI, or Settings → Accounts & keys in the desktop app.

You pay for a subscription (Claude, ChatGPT, Copilot, and others)Link to this section

Run /login inside Empryo. A list of accounts opens. Pick yours and follow the browser or device-code prompt. You can also name the account directly:

/login claude      # Claude Pro or Max
/login codex       # ChatGPT
/login copilot     # GitHub Copilot

After you sign in, press Ctrl+L and pick a model from that account. See Subscriptions for every account you can sign in to, and the risks involved.

You want to run models locallyLink to this section

Start Ollama or LM Studio. Empryo finds them automatically. No key is needed.

Cloud providersLink to this section

Get a key from the provider's site, then save it:

Gateways: one key, many modelsLink to this section

A gateway resells models from many vendors behind one key and one bill.

Coding plansLink to this section

Some vendors sell a monthly plan for coding tools, with its own endpoint. Each one needs only a key.

Empryo reads the model list, context sizes, and prices for these plans from the models.dev catalog. When a vendor adds a model, you can pick it without waiting for an Empryo update. A plan only appears in the model picker after you set its key.

Save the key like any other:

empryo --set-key zai-coding-plan ...
empryo --set-key zhipu-coding-plan ...
empryo --set-key alibaba-coding-plan ...

Z.AI, Alibaba (both plans), Xiaomi, and Kilo can also be added with /login. For Alibaba and Xiaomi, /login tests your key against each of the vendor's regional servers and remembers the one that works, so you do not need to know which region you bought. See Subscriptions.

If you already export the vendor's usual environment variable, Empryo uses it: ZAI_API_KEY, ZHIPU_API_KEY, ALIBABA_CODING_PLAN_API_KEY, ALIBABA_TOKEN_PLAN_API_KEY, XIAOMI_API_KEY, KILO_API_KEY, SILICONFLOW_API_KEY, SYNTHETIC_API_KEY.

Local models (free)Link to this section

Enterprise and subscriptionsLink to this section

LLM GatewayLink to this section

LLM Gateway gives you one key for models from many vendors, including Claude, GPT, Gemini, Grok, Llama, and Qwen, with cost tracking. It also offers DevPass, a fixed-price monthly subscription.

1. Type /login inside Empryo and choose LLM Gateway. Your browser opens, creates a key, and sends it back to Empryo. You can also get a key at llmgateway.io and save it yourself:

empryo --set-key llmgateway sk-...

1. Start Empryo, press Ctrl+L, and pick a model.

The model picker shows how you are paying: Pay-as-you-go · account credits or DevPass · subscription. To switch between them, change the key in /keys.

Choose which vendor serves a modelLink to this section

LLM Gateway can get the same model from several vendors, and their prices and output limits differ. By default the gateway picks the vendor for you. With a pay-as-you-go key, you can choose it yourself:

• Terminal UI: in the model picker, a model with several vendors shows ⇌ N. Press Ctrl+R on it. You get Auto — gateway decides first, then each vendor from cheapest to most expensive, with its price and output limit. After you pick one, the row reads via <vendor>.
• Desktop app: click the ⇌ button on a gateway model to see the same list.
• Headless: put the vendor in the model id.

empryo --headless "fix the tests" --model llmgateway/anthropic/claude-opus-5
empryo --headless "fix the tests" --model llmgateway/deepinfra/deepseek-v4-pro

llmgateway/<vendor>/<model> uses that vendor. llmgateway/<model> lets the gateway decide.

It is still the same model. Favorites, recent models, pricing, and context size all carry over. The vendor you choose applies to that tab only, so two tabs can use one model through different vendors. Choosing a vendor only works with pay-as-you-go keys. DevPass always lets the gateway decide.

OpenRouterLink to this section

OpenRouter offers hundreds of models behind one key, including both frontier and open-weight models.

The easiest way to connect is to sign in with your browser. There is no key to copy:

empryo
# then: /login openrouter

Or save a key yourself:

1. Get one at openrouter.ai/keys.
2. Save it:

empryo --set-key openrouter sk-or-...

1. Start Empryo, press Ctrl+L, and pick a model.

Your OpenRouter credit balance appears in /usage and in the desktop app's usage page, next to what you have spent.

Anything elseLink to this section

Any API that follows the OpenAI format works as a custom provider. You add one block to your config, with no code. Save the key under the id you gave the provider:

empryo --set-key together sk-...     # `id` from your providers[] entry

Run the command again with a second key and Empryo keeps both. When one account runs out of quota, Empryo moves to the next. See several accounts on one provider.

Switching modelsLink to this section

Inside Empryo, press Ctrl+L to pick a provider and model. To use different models for different jobs, such as a cheap model for searching and a strong one for writing code, use /router.

From the command line:

empryo --list-providers              # see what's configured
empryo --list-models anthropic       # see available models for one provider