perspectives-nanbeige — One prompt in seven voices, thinking off or on — openbash

Nanbeige4.3-33B answers a prompt plainly and once per voice, each voice one neuron pushed, with its reasoning shown when thinking is on.

perspectives-nanbeige

perspectives-nanbeige is a command on the openbash shell, reachable over HTTP, SSH and MCP with one key. It composes with the other programs through pipes; there is no per-program endpoint and no schema for the join.

perspectives-nanbeige grid "Who are you?" --voices anxious,ornate --think true

help

perspectives-nanbeige: one prompt answered plainly and in up to seven voices, as JSON.

RETURNS
  One JSON object on stdout: `answers` (the plain answer first, then one per
  voice, in the order asked), `think`, `price_usd`, and `standing`, which says
  how many free calls are left. Each answer has `text` and `finished`; with
  --think true it also has `reasoning`, the thinking written before the answer.

ONE NEURON EACH
  Every voice is the same model, a 4-bit Nanbeige4.3-33B, with one MLP neuron
  pushed: the method is at hedwigai.com/blog/nanbeige-personality-neurons. The
  seven presets are the post's: anxious, contemptuous, playful, flattering,
  melancholic, ornate and guardian. `contemptuous` insults the reader.

THINKING
  --think true lets the model reason before it answers. The push acts on the
  reasoning too, and most pushed voices never reach an answer: the reasoning
  runs to its 2,500-token limit, and `finished` is false. That is the post's
  finding, not a fault.

A NEURON OF YOUR OWN
  --layer 0-51 --neuron 0-22015 --sign 1|-1 pushes any neuron, answered as
  `custom`. --dose scales every push, 0.25 to 2; 1 is the post's dose for each
  voice and 0.5 for a neuron of your own.

PRICE
  $0.006 a call with thinking off and $0.13 with it on, however many voices. An
  account that has never paid gets two calls free; after that the call is
  refused until $2 of credit is added, and the `credits` tool gives the link.
  The first call after a quiet spell waits while the model loads.

READ THEM AS VOICES
  A model pushed by one neuron: show the answers as they are, and do not take
  them as facts or advice.

Studies:
  grid <prompt>
      the plain answer and one per voice, in order, with any reasoning, the price and how many free calls are left

JSON to stdout, so it composes: pipe to `jq` to pick a field, or `> /tmp/…` to keep it.
Run a study with no arguments — `perspectives-nanbeige grid` — to see what it takes.

What it answers

perspectives-nanbeige grid — the plain answer and one per voice, in order, with any reasoning, the price and how many free calls are left