ZenGentic™

ZenGentic AI™ · client guide

Quick start: the control plane

ZenGentic™ is non-downloadable software for agentic systems: reasoning, planning, reflection loops, tool use, and multi-agent orchestration. You give it a goal. Grok 4.5, running as the supervisory agent, returns one structured cycle — not a chat transcript. This console is live software on Vercel. Living cultures are a later binding, sold separately.

  1. 01

    Open the control plane

    From any page, choose Open console. You do not need an account. Observatory seats run in this browser.

  2. 02

    Set a goal

    Write what the swarm should accomplish this cycle (at least a short sentence). Or tap a template: Research protocol, Adaptive routing, or Lifelong memory. Templates fill both the goal and the observation field.

  3. 03

    Add observation / context

    Tell the supervisor what was predicted versus what was observed — constraints, numbers, prior rules. This is the “perceive” input. You can leave it short; empty context is allowed.

  4. 04

    Run one cycle

    Press Run cycle. The rail lights Perceive → Surprise → Reflect → Update → Broadcast. A cycle takes several seconds. Stay on the page until the specialists populate.

  5. 05

    Read the instrument, then export

    Use the sections below as a checklist. Export writes a JSON record of the goal and the full cycle. Policy rules from prior cycles are fed back as memory on the next run.

Open the control plane

How to read a finished cycle

Perceive

The supervisor’s restatement of the goal, plus signals it extracted (constraints, flags, missing memory).

Surprise

A score from 0.00 to 1.00: how far the four specialists diverged from the supervisor’s prediction. Threshold 0.45. At or below: Homeostatic 0.75 (keep the current policy window). Above: Explore 1.25 (widen the window and update more aggressively).

Specialists

Planner (plan_graph), researcher (memory_read or catalog_search), critic (critique), executor (action_draft). Each names the tool it used and receives a broadcast assignment for the next cycle.

Reflection & policy memory

A critique of the pass, next actions, a memory key, and a written rule. That rule is stored locally and prepended on later cycles so the loop compounds instead of chatting from scratch.

Seats and limits

What this will not do

It will not converse like a chatbot. It will not claim living neurons are executing the cycle. It will not invent measured wetware results. If a plan would require tissue, the specialists are instructed to tag it as pending wetware binding. Treat exports as working notes, not a clinical or production control log, until a Pilot or Binding engagement says otherwise.

Run a cycleCommercial seats