The language-model route that would answer

Reports which route llmask and llmchat would take right now, without sending a prompt anywhere. Use it to check, before asking a question, that the question would go to the server you expect (your own endpoint, a provider whose key is set, or a local Ollama server) and which model it would use.

Usage

llmrte(route = NULL)

Arguments

route

Character scalar or NULL (the default). NULL uses the saved route, else the automatic order; "auto" uses the automatic order and ignores a saved route; "own", "ollama" or a provider id from the id column of llmprvd (such as "anthropic") asks about that one route only.

Value

NULL when no route is set up (or the named one is not available), else a list of four elements: route (character, the id that route = takes: "own", "ollama" or a provider id), name (character, a readable label such as "own endpoint", "local Ollama" or "Anthropic"), model (character, or NULL when no model is chosen) and base (character, the base address requests go to, without a trailing /v1 for your own endpoint).

Details

The automatic order is: an OpenAI-compatible endpoint of your own (MORIE_LLM_BASE_URL, or own.url saved with rmoriebricklayer config set); then every provider of llmprvd whose key is set, in the order of that table; then a local Ollama server, at OLLAMA_HOST or by default at http://localhost:11434, when it answers with at least one pulled model or OLLAMA_MODEL is set. A saved route replaces the automatic order when route = NULL; it comes from MORIE_LLM_ROUTE or from the route saved in the settings file.

Nothing is sent to a provider. The only network contact is the Ollama probe, a GET /api/tags to the Ollama address with a 2 second timeout; OLLAMA_HOST=off skips it. Your own endpoint is not probed.

Errors: route = "hosted" (the hosted MORIE tier is retired); a route that is not one of the accepted names; a provider route named explicitly with no key set; a MORIE_LLM_BASE_URL that is not https to a public address or plain http to the loopback host or a literal private LAN address. When automatic resolution skips a route that is half set up (a model or key without an address, an Ollama that does not answer) a message says so.

See also

llmstat for every route at once, llmprvd for the provider ids.

Examples

# keep the examples away from your real settings and from the network
old <- Sys.getenv(
  c(
    "XDG_CONFIG_HOME",
    "OLLAMA_HOST",
    "MORIE_LLM_BASE_URL",
    "MORIE_LLM_MODEL"
  ),
  unset = NA
)
Sys.setenv(XDG_CONFIG_HOME = tempfile(), OLLAMA_HOST = "off")

# Ollama switched off: that route is not available
llmrte("ollama")
#> NULL

# an OpenAI-compatible server of your own on this machine (not contacted
# here)
Sys.setenv(
  MORIE_LLM_BASE_URL = "http://localhost:8080/v1",
  MORIE_LLM_MODEL = "llama3"
)
str(llmrte())
#> List of 4
#>  $ route: chr "own"
#>  $ name : chr "own endpoint"
#>  $ model: chr "llama3"
#>  $ base : chr "http://localhost:8080"

# the retired hosted tier, and a route name that does not exist
try(llmrte("hosted"))
#> Error : the hosted MORIE language-model tier is retired: bring your own key (a provider's variable such as ANTHROPIC_API_KEY or OPENAI_API_KEY, or `rmoriebricklayer config set <provider>.key`) or run a model locally (Ollama); `rmoriebricklayer config setup` walks through it
try(llmrte("nope"))
#> Error : llmrte: `route` must be one of "auto", "own", "openai", "anthropic", "gemini", "openrouter", "groq", "together", "fireworks", "deepseek", "mistral", "xai", "cerebras", "perplexity", "huggingface", "opencode", "cohere", "nvidia", "ollama", not "nope"

for (v in names(old)) {
  if (is.na(old[[v]])) {
    Sys.unsetenv(v)
  } else {
    do.call(Sys.setenv, as.list(old[v]))
  }
}