Navigation
ELOWEN · Documentation

Last updated: October 27, 2026

Browse documentation · Add an API-key provider
AI models and providers

Add an API-key provider

Connect any OpenAI-compatible address or an Anthropic key, then say which of its models the instance offers.

Add it in Settings

  1. Open Settings → <assistant name> AI → AI providers and choose Add provider.
  2. Enter a Name.
  3. Choose the type: OpenAI-compatible or Anthropic API.
  4. Enter the API address. For an OpenAI-compatible address, include its version segment, for example https://ai.example.com/v1.
  5. Paste the API key.
  6. Enter model IDs, one per line, or leave the list empty and let Chetty load them (see below).
  7. Save. The provider appears in the list, and its models can be granted to accounts.

The setup wizard and the CLI provider picker offer presets that fill in the address; you still enter the key. The presets are OpenAI, Anthropic (Claude), OpenRouter, Google Gemini, xAI (Grok), DeepSeek, Groq, Mistral, Together AI, Fireworks AI, Cerebras, Perplexity, DeepInfra, Moonshot (Kimi), Z.AI (GLM), NVIDIA NIM, Hugging Face, Baseten, Ollama Cloud and Ollama (local).

Note: A connection alone does not let every account use those models. Connect a provider and offer models covers the grants.

Enter the models

An empty model list makes Chetty discover the models at <base-url>/models, and a successful discovery can also add context-window information. As soon as you type model IDs, your list controls which models are offered.

If an address has no /models route, enter the model IDs by hand.

Choose the wire format

Wire API chooses the API the provider speaks: Auto, Responses or Chat Completions.

  • Auto picks the newest supported API for each model and remembers the result. OpenAI and known account catalogs use their native format.
  • For another compatible address, the first use starts a small paid check in the background. Your conversation begins at once on the default format; a detected change takes effect with the next message and appears in the conversation, and a failed check keeps the current default.
  • Choose Responses or Chat Completions to fix the format for the whole provider, or pin one model's format in that model's settings.

To see each model's effective API, open the provider's ⋯ → Settings → General → Models → Manage; the API sits under the model's name. With Auto and a remembered result, the Reset API icon in the model's limits forgets that detection.

Set the temperature

Temperature is optional and accepts 0 through 2. Leave it empty to let the model use its own default.

Note: Some models reject any other value, and Kimi K3 and Claude Opus 4.7 and newer accept only their default.

The rest of the network settings sit under Advanced: the compatibility options for optional Chat Completions features, and the tool-execution switches this provider supports.

Gemini and Ollama

Gemini uses the Google OpenAI-compatible address https://generativelanguage.googleapis.com/v1beta/openai. Chetty does not speak the native Google API, so Gemini models that an Opencode address offers only through that API cannot be used through Chetty.

Ollama (local) uses the built-in address http://127.0.0.1:11434/v1 and needs no key. The setup wizard can install Ollama; a manual setup must already be running on the Chetty host.