LorcaDocs

AI providers

Connect an API-key provider or subscription once, and every Runner on your account can use it.

Bots think with your own AI accounts. Connect a provider once, on any paired device, and it reaches every other device encrypted with your account key.

ProviderHow you connectDefault model
DeepSeekAPI keydeepseek-flash
AnthropicAPI keyclaude-opus-5
OpenCode ZenAPI key, pay as you godeepseek-v4.1-flash
OpenCode GoAPI key and Go subscriptionglm-5.3-flash
ChatGPTSign in with your ChatGPT subscriptiongpt-6.1-sol
GrokSign in with SuperGrok or X Premium+grok-4.7

DeepSeek, Anthropic, ChatGPT, and Grok search the web on the provider's side. Grok also searches X.

Connect

In the desktop or phone app, open Settings › Providers, select a provider, and choose Connect.

  • API key: paste the key. Lorca checks it against the provider before saving.
  • Sign-in: your browser opens the provider's own sign-in page. Lorca never sees your password.

From a terminal:

lorca provider set deepseek
lorca provider set anthropic
lorca provider set opencode
lorca provider set opencode-go
lorca provider set chatgpt
lorca provider list

Custom API endpoint

For a proxy or a compatible server, fill in Base URL when you connect any API-key provider. The key is checked against that endpoint.

Custom providers

Any server that speaks OpenAI's Chat Completions or Responses API, or Anthropic's Messages API, can be a provider: a gateway such as OpenRouter, another vendor's API, or a model server on your own network such as Ollama, LM Studio, or vLLM. Open Settings › Providers, choose Add Provider…, and pick the service: OpenAI, OpenRouter, Gemini, Groq, Together AI, Ollama, or LM Studio fill in the API and base URL for you, and Other Server… starts empty.

  • Name: what the apps call it.
  • API: OpenAI Chat Completions, OpenAI Responses, or Anthropic Messages.
  • API base URL: the root the API's paths go under, such as https://openrouter.ai/api/v1 or http://localhost:11434/v1. The sheet shows the full URL Lorca will call.
  • API key: leave it empty for a server that takes none.
  • Models: as you fill in the server, its models appear with their context window and whether they see images. Check the ones bots can use and pick the default. For a model the server doesn't list, type its ID in the search field and add it.

Lorca checks the server again when you save, which checks the key. A custom provider syncs to your other devices like any credential, and its bots run on whichever Runner they are assigned to, so a model server on your network has to be reachable from that Runner. Spending shows as $0, since Lorca does not know what the server charges.

From a terminal:

lorca provider add "OpenRouter" https://openrouter.ai/api/v1 --model anthropic/claude-sonnet-5 --api-key-stdin
lorca provider add "Ollama" http://localhost:11434/v1
lorca provider add "Claude proxy" https://proxy.example.com/v1 --api messages --model claude-opus-5 --api-key-stdin

Decision APIs

A decision model answers typed questions with probabilities instead of writing text. Auto-review can run one, and bots can't. Add one under Add Provider…: OpenRouter Decisions, OpenAI Decisions, and TypeSafe fill it in. Under Other Server…, pick the API:

  • System One: TypeSafe's API, which OpenRouter, OpenCode Zen, and LLM Gateway serve too.
  • OpenAI Decisions: OpenAI's Decisions API, also served by Vercel's AI Gateway.

For these two, the URL is the endpoint itself, such as https://openrouter.ai/api/alpha/decisions. A URL without one gets the API's usual path (/systemone or /decisions). The model list shows the server's decision models. A decision provider doesn't appear in the bots' provider pickers. Pick it in Settings › Auto-review, and its model under Review Models.

lorca provider add "TypeSafe" https://api.typesafe.ai/v1/systemone --api system-one --api-key-stdin

Review models

Settings › Providers › Review Models has each connected provider's review model, the one Auto-review runs on it. By default it's a small, fast model; pick any of the provider's models, decision models included.

Pick a model per bot

Each bot chooses its provider, model, and thinking level in the New Bot sheet, under Runs with in the desktop-app inspector, or in the phone's bot Details. A bot with no model set uses the provider's default.

If a bot's provider is not connected, its turn ends with a note that says to connect it in Settings.

Disconnect

Disconnect removes the credential from every device. For Grok it also revokes the sign-in. Delete removes a custom provider from every device; lorca provider remove does the same from a terminal, with the kind lorca provider list shows, such as custom:openrouter.

Cost

DeepSeek, Anthropic, and OpenCode Zen bill by usage; OpenCode Go applies usage to its subscription limits. The inspector shows what each chat has spent. ChatGPT and Grok sign-ins use your subscription, and the figure shown is what the work would cost at API rates.

On this page