Connect a model provider

Excalibur needs a model provider before it can answer requests. Have the provider’s model identifier ready. For a hosted service, you also need its API credentials. For a local Ollama server, install and start that server separately and make sure the selected model is available there.

Try a local provider

Add this table to your user config.toml, replacing the model placeholder:

[models]
provider = "ollama"
model = "YOUR_MODEL_ID"

On Linux and macOS the default file is ~/.config/excalibur/config.toml, or the equivalent under XDG_CONFIG_HOME. On Windows it is under %APPDATA%\Excalibur. Ollama uses its local default endpoint unless you set base_url. The Ollama adapter does not require an API key.

After saving the file, run the first request in Quick start. Restart your daemon after changing settings used by an already running daemon.

Use a hosted service

The API provider adapters include anthropic, gemini, openai, deepseek, and openrouter. Use the provider identifier and an actual model identifier from your account. A provider name alone does not select a model for you.

The standard credential environment variables are:

Provider	Credential variable
Anthropic	ANTHROPIC_API_KEY
Gemini	GEMINI_API_KEY or GOOGLE_API_KEY
OpenAI	OPENAI_API_KEY
DeepSeek	DEEPSEEK_API_KEY
OpenRouter	OPENROUTER_API_KEY

Set the credential through your terminal or service’s secret configuration. Do not paste it into chat or commit it in a project configuration file. The process running Excalibur must inherit the variable.

A compatible endpoint can use generic with an explicit base_url. Account sign-in is also available through Settings → Connections → ChatGPT / Codex account; connecting that account and selecting model routing are separate steps. See Configuration for environment overrides and vault references.