Skip to contents

Call Local Ollama or OpenAI-Compatible LLM Endpoint

Source: R/llm.R
sv_call_llm.Rd

Sends prompt requests to local Ollama (http://localhost:11434) or OpenAI-compatible endpoints via httr2. Supports model selection, customizable endpoints, API key authentication, and clean error reporting for network or HTTP status failures.

Usage

sv_call_llm(
  prompt,
  backend = c("ollama", "openai"),
  model = NULL,
  endpoint = NULL,
  api_key = NULL,
  timeout = 60,
  extract_code = FALSE,
  resolve_model = TRUE,
  ...
)

Arguments

prompt

A character string or sideview_prompt object from sv_build_prompt.

backend

Character string: either "ollama" or "openai".

model

Model name (e.g., "llama3.2", "llama3", "gpt-4o-mini").

endpoint

Base URL or full chat endpoint for the API.

api_key

API key (for OpenAI or authenticated endpoints).

timeout

Request timeout in seconds. Default 60.

extract_code

Logical; if TRUE, runs sv_extract_code() on the response text. Default FALSE.

resolve_model

Logical; for Ollama, if an untagged model such as "llama3.2" is not installed, retry once with an installed tag of it (e.g. "llama3.2:3b"). Default TRUE.

...

Additional options added to the request body (e.g. temperature). For Ollama, think = FALSE is sent unless you pass think or set options(sideview.llm_think = TRUE), so reasoning models answer promptly.

Value

Character string containing the LLM response text (or extracted code if extract_code = TRUE).