Custom endpoint

Olyvr works with any service that offers the OpenAI chat completions API with streaming and tool calling. That includes most hosted providers, AI gateways, and self-hosted servers such as llama.cpp, vLLM or Ollama’s OpenAI-compatible mode.

Configure in the app

1. Open Settings → Model access and expand Endpoint & advanced.

2. Enter the base URL, ending in /v1 for most providers, for example https://api.example.com/v1.

3. Enter the API key if the endpoint needs one, and save.

4. In a chat, open the composer’s model picker. Olyvr lists the models the endpoint reports at <base URL>/models.

The key is encrypted on disk and is never shown again after you save it. To change it, paste a new key; to remove it, clear the key.

Caution

The base URL must start with http:// or https:// and must not contain a username or password. Put credentials in the API key field.

Configure with environment variables

For managed machines or scripted setups, set these before starting Olyvr. They take priority over the values saved in Settings and are never written to disk:

Variable	Example
ATLAS_MODEL_BASE	https://api.example.com/v1
ATLAS_MODEL_KEY	sk-…
ATLAS_MODEL	gpt-5.5

Choosing the model

You pick the model per conversation in the composer. When there is no choice yet, Olyvr prefers a model whose name starts with gpt- or claude-. Endpoints on your own machine (localhost or 127.0.0.1) are labelled as local.

Troubleshooting

Message	Meaning
“Configure a model endpoint in Settings → Model access…”	No endpoint is set. Add one, or use a subscription.
“Model endpoint returned HTTP 401/403…”	The API key is wrong or lacks access to that model.
“Model endpoint returned HTTP 400/404…”	The model name is wrong, or the endpoint doesn’t support streaming or tool calls.
“The model returned no answer or tool calls.”	The model replied with nothing usable. Try again or choose a more capable model.

Requests time out after two minutes.