Provide a model code string and its compile error message to be refined

Sends a string containing model code and error message to a Dify Workflow or LLM API for correction.

Usage

refine_model_code(
  model_code,
  error_message,
  context = NULL,
  reuse_context = FALSE,
  service,
  api_key,
  api_upload = NULL,
  api_chat = NULL,
  user_id = "mrgsolve_translator",
  model_gemini = "gemini-3-flash-preview",
  model_openai = "gpt-5-mini",
  model_anthropic = "claude-haiku-4-5-20251001",
  model_openrouter = "arcee-ai/trinity-large-preview:free",
  model_openai_compatible = "gpt-5-mini",
  model_deepseek = "deepseek-reasoner",
  model_apollo = "gpt-5.2",
  model_azure = "gpt-5.2",
  model_aws = "anthropic.claude-sonnet-4-6",
  progress_bar = 0.4,
  max_retries = 2,
  display_info = TRUE,
  temperature = 0,
  seed = 42,
  attempt = 1,
  deep_pdfscan = FALSE,
  force_parse = FALSE,
  system_prompt,
  long_user_prompt,
  short_user_prompt,
  internal_version,
  feedback_success = FALSE,
  debug = TRUE
)

Arguments

model_code: Input model code
error_message: Input compilation error message
context: Argument to hold chat_obj or conversation_id to keep same chat session
reuse_context: Set to TRUE to re-use same conversation_id or chat_obj, more costly in terms of time and tokens
service: Choice of "PROD", "EXP", "Gemini", "OpenAI", "Claude", "OpenRouter", "OpenAI-Compatible", "DeepSeek", "Apollo", "Azure OpenAI", "AWS Bedrock"
api_key: API Key, recommended to store it as env var called "ANTHROPIC_API_KEY" etc (Dify only)
api_upload: API URL for uploading of files (Dify requires a 2-step process)
api_chat: API URL for chat messages, required for OpenAI-compatible
user_id: user id for the request
model_gemini: Model to be used when calling Gemini API
model_openai: Model to be used when calling OpenAI API
model_anthropic: Model to be used when calling Anthropic API
model_openrouter: Model to be used when calling OpenRouter
model_openai_compatible: Model to be used when calling OpenAI-compatible API
model_deepseek: Model to be used when calling DeepSeek
model_apollo: Model to be used when calling Apollo LLM
model_azure: Model to be used when calling Azure OpenAI
model_aws: Model to be used when calling AWS Bedrock
progress_bar: Starting point of Shiny progress bar, relevant for 2nd try
max_retries: Required to calculate a more accurate progress bar
display_info: Set to TRUE to show how much time/tokens in Shiny UI when job is finished
temperature: Goes from 0 to 1, where 0 is deterministic, if not reusing context
seed: seed number for LLMs, if not reusing context
attempt: current attempt number
deep_pdfscan: Uses Vision to extract image data (BI only)
force_parse: Force reparsing (BI only)
system_prompt: Character. System prompt
long_user_prompt: Character. Long user prompt
short_user_prompt: Character. Short user prompt
internal_version: Logical. Only relevant for BI
feedback_success: Logical. Only relevant for BI, and if reuse_context = TRUE
debug: Displays debug messages

Value

a named list of "answer", "conversation_id", "chat_obj"