hrdr-0.4.2 is not a library.
hrdr — herder: an agentic coding harness.
No subcommand launches the interactive TUI. hrdr run <task> runs a single
turn headlessly, streaming to stdout (scriptable, pipeable).
hrdr models lists available models from the configured endpoint.
hrdr talks to any running OpenAI-compatible endpoint; name the model you want
as provider://model (--model chatgpt://gpt-5.5). The endpoint is a property
of the PROVIDER — a built-in preset, or a [providers.<name>] table in
config.toml — so a server of your own is a provider you define, not a flag. It
does not manage a model server — start your own (infr, llama.cpp, vLLM, …) or
point at a hosted provider.