Agent-assisted reproducibility-bundle help

Asks a language model to plan a brick-proof, reproducible data bundle, through llmask: your own key, your own endpoint or a local Ollama server, with the packages' help at hand (the model reads the help pages of the functions it proposes before it answers). Use it to get a first draft of the steps and R code for packaging an analysis: finding the portal's download URL, fetching with a Wayback fallback, recording hashes and a manifest, writing a synthetic stand-in, then packaging, signing and verifying the bundle.

Usage

agentbndl(req, model = NULL, route = NULL)

Arguments

req

Character scalar describing the bundle task.

model

Optional model id, a string; NULL uses the route's model (rmbl models lists them).

route

NULL (the saved route, else the first that is set up), "auto", "own", "ollama" or a provider id such as "anthropic" or "openai".

Value

Character scalar: the answer, or a message saying what to set up when no route answers.

Details

The request is sent with a fixed system prompt that describes that task and asks for a short plan and R code. The route is chosen as in llmask(); when no route is set up, the function returns a message saying what to set up (a provider key, a local Ollama or your own endpoint) instead of an error, and makes no network request. When a route is set up, the request and the prompt are sent to that provider or server. The answer is model output: check the code it proposes against the help pages before running it.

Errors: req not a single non-empty string; route not one of the names listed; route = "hosted" (the hosted tier is retired); and errors raised by llmask().

Examples

if (FALSE) { # interactive()
# Needs a language model: a provider key (e.g. ANTHROPIC_API_KEY), your own
# endpoint or a local Ollama.
agentbndl("scaffold a bundle for analysis.R from the Toronto CKAN data")
agentbndl("sign and verify my capsule", route = "ollama")
}
# With every route switched off (no key in the environment and no saved
# settings: the settings folder is pointed at an empty one) the call returns
# a setup hint, not an error, and reaches no network -- safe to run anywhere:
keys <- c(
  "OLLAMA_HOST",
  "MORIE_LLM_BASE_URL",
  "XDG_CONFIG_HOME",
  llmprvd()$key_env,
  llmprvd()$alt_env[nzchar(llmprvd()$alt_env)]
)
old <- Sys.getenv(keys, unset = NA)
Sys.setenv(
  OLLAMA_HOST = "off",
  MORIE_LLM_BASE_URL = "off",
  XDG_CONFIG_HOME = tempfile()
)
Sys.unsetenv(setdiff(
  keys,
  c("OLLAMA_HOST", "MORIE_LLM_BASE_URL", "XDG_CONFIG_HOME")
))
cat(agentbndl("hello"))
#> No language-model route is set up on this machine: bring your own key, or run a model locally:
#>   a provider's key: set its variable (OPENAI_API_KEY, ANTHROPIC_API_KEY, GEMINI_API_KEY, OPENROUTER_API_KEY, ...) or `rmoriebricklayer config set PROVIDER.key`; providers: openai, anthropic, gemini, openrouter, groq, together, fireworks, deepseek, mistral, xai, cerebras, perplexity, huggingface, opencode, cohere, nvidia
#>   local Ollama: OLLAMA_HOST=off
#>   own endpoint: `rmoriebricklayer config set own.url URL` (and own.model, own.key) for any OpenAI-compatible server (LM Studio, vLLM, llama.cpp, a company gateway)
#> `rmoriebricklayer config setup` walks through it.
for (v in keys) {
  if (is.na(old[[v]])) {
    Sys.unsetenv(v)
  } else {
    do.call(Sys.setenv, as.list(old[v]))
  }
}

# argument errors come first
try(agentbndl(""))
#> Error : agentbndl: `req` must be a single non-empty string
try(agentbndl("plan a bundle", route = "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