
Build the LLM client used for dimension scoring
liminal_chat.RdCreates an ellmer chat client for the OpenAI-compatible Panama
endpoint. All configuration comes from the environment; the API key
is never stored in source. Temperature is fixed at 0 for scoring,
because scores must be comparable across business cases.
Usage
liminal_chat(
model = Sys.getenv("LIMINAL2_MODEL", "glm-5.3"),
base_url = Sys.getenv("PANAMA_BASE_URL", "https://panama.ixpantia.ixops.ai/v1"),
api_key = Sys.getenv("PANAMA_API_KEY")
)Arguments
- model
The model name. Defaults to the
LIMINAL2_MODELenvironment variable, falling back to"glm-5.3".- base_url
The endpoint URL. Defaults to the
PANAMA_BASE_URLenvironment variable, falling back to the Panama endpoint.- api_key
The bearer key. Defaults to the
PANAMA_API_KEYenvironment variable; an informative error is raised when unset.
Details
The chat object carries the configuration and the provenance of the
client; the scoring calls themselves are issued by the package's
httr2 transport (see assess_liminal_space()).
Examples
# \donttest{
if (nzchar(Sys.getenv("PANAMA_API_KEY"))) {
chat <- liminal2::liminal_chat()
}
# }