Hands vs brain

Do AI agents need a brain to use a terminal? · Chefe 2668 · 21 Aug 2026

Ideia central

A terminal is hands. An LLM is the brain. An AI agent is both. You do not need a brain to open a window or run a named script. You do need a brain for agents to think, write tools, and spawn children at speed. The brain is not inside ParviClaw Core. Core only opens the hands. The brain is one thing the customer already has: a cloud API or a local model on their machine. Not both required. 65 empty windows with no brain = fast at nothing.

Terminal equals hands. LLM equals brain (cloud API or local). Agent uses the hands. Range prefers local. Defenders pick either.

Terminal = hands

A windowcan type. cannot think.
No brain needed to open ita human or a script can use it

LLM = brain

Thinks, decides, writes toolsthen uses the hands
Need ONEcloud API or local model

The agent’s brain does the thinking. The terminals do not “do it all” by themselves.

O que foi entregue

  1. Your “go a” — engine default is now 72 live terminals (65+ works; 65th is not the last slot). Paid pro license was 50 (would have blocked you) → 72. Tests pass.
  2. The confusion, in one table
    QuestionAnswer
    Need a brain just to use a terminal?No. Hands only.
    Need a brain for AI agents to do the job?Yes. Brain thinks; hands type.
    Must they buy a cloud API from day one?No. Local LLM is enough.
    Must they run a local LLM?No. A cloud API is enough.
    Must they have one brain before agents work?Yes — API or local. One.
  3. The two products
    Usual brainWhy
    ParviRange (military)Local LLM on their machine (air-gap). Cloud API only if their rules allow.Their data often cannot leave the building.
    ParviDefenders (commercial)Cloud API, local LLMs, or both mixed (hybrid) — Chefe 2677.Office keys and on-machine models can run together.

O que mudou no meio do caminho

We did not turn on “refuse to start until they name a brain” (that was option b). You said go a only, and you were confused about the brain — so we explain first. House still uses our hub. Customer SKUs still need a later install question once you say yes.

O que falta

Only you — one yes/no:

For Defenders and Range, should install refuse to start AI waves until they name one brain (their cloud API or their local LLM)? Terminals can still open empty for a human. Agents cannot think without that name.

If yes, that is option b, now clearer. If no, they can start and we warn them the swarm is empty until they plug a brain.