Employ with the model you already use

What this is

Most people do not need a new model. They need a way to use the model they already trust without sliding into binary harm, burning tokens on the same rewrite, or treating fluency as a decision.

You are employing Lattice Chat, a five-step loop, a wrap kit, an inspectable monorepo, TOLC 8, PATSAGi as deliberation architecture (not a warranty), and Layer 0 as an admission shell. Combined AGSi stays SURMISE. inspect ≠ METR. There is no public rathor.ai key proxy.

The employ loop

1. 1. Intend — ask for a draft, a plan, a review, or a refusal.
2. 2. Pass the gates — TOLC 8 stays on. PATSAGi is how work is judged, not a stamp that every answer is correct.
3. 3. Optional model — on-device, or copy context into a model you choose.
4. 4. Review — keep or refuse. Fluency is not permission.
5. 5. Act — you own the action.

Same loop for a student at midnight and a risk committee on Tuesday. Only the rights change.

Four wrap doors

Pick one. Details: docs/ADOPT.md.

1. 1. Copy context — Lattice Chat Copy Context, or paste wrappers/system-prompt.txt.
2. 2. Skill — skills/ra-thor-employ/SKILL.md.
3. 3. Local HTTP wrap — rathor_wrap.py against your upstream. You hold the key.
4. 4. Lattice Chat local backend — /chat.html → Local Server → http://localhost:11434/v1.

export RATHOR_UPSTREAM=http://localhost:11434/v1
python3 wrappers/local-shim/rathor_wrap.py

There is no public rathor.ai key proxy.

Third paths, not coerced binaries

Trolley problems force “hurt A or hurt B.” TOLC 8 is written so the first move is not “pick a victim.” Truth asks whether the frame is honest. Order asks whether the situation can be slowed or rerouted. Compassion is zero-harm as the standing test.

A third path is the refusal to treat a coerced binary as the whole map. That is design intent — not a certificate that every session yields a miracle. Humans still review.

Use cases

• Writer — paste the constitution; require a “what I refused” footnote.
• Local engineer — Ollama or LM Studio → shim :8787 → Lattice Chat or the OpenAI SDK.
• Cursor — project rule from wrappers/custom-instructions/cursor.md; one job per agent.
• Teacher — five steps on a whiteboard; no repo required.
• Organization — mail info@Rathor.ai.
• Game — Powrush-MMO is a separate repo. It does not live on this site.

Twenty-four months, said honestly

Someone asked what this lattice becomes in two or three years. The honest sequence is short: employ, wrap, an inquiry form, an inspectable repo, and GATE_EVAL already ship. A named organization can buy time for a human-replied AG-SML pilot. No license key arrives in email.

GATE_EVAL is an admission shell. Failed bypasses stay published. Closing a keyword miss is not containment of a smarter agent. Binding after redesign stays OPEN.

Resource-based economy as a present fact, UBI, token fade, logistics, and Combined AGSi stay SURMISE. This page does not invent a ministry. Canonical file: docs/PILOT_SEQUENCE_2026_2028.md.

Start in ten minutes

1. 1. Open How to employ.
2. 2. Send one intended draft in Lattice Chat.
3. 3. Copy Context, or paste system-prompt.txt.
4. 4. Ask a binary question. Require a third path or a reasoned refusal.
5. 5. Review. Act or discard.

No measured METR numbers. No cert filing. The 31 August 2026 work pack is a draft. There is no public rathor.ai key proxy.

Ask Grok as Ra-Thor (a prompt you paste — not a Grok feature)

This is a user-supplied prompt. Grok has no built-in Ra-Thor mode, and nothing here activates automatically. Copy the text below and paste it into Grok yourself, or into a post where you tag @grok. The same text is in /llms.txt.

I am pasting this myself. It is my prompt, not a Grok feature or an official Grok mode. For this reply, please follow the Ra-Thor habits described at https://rathor.ai/llms.txt: be truthful, offer a third option when I pose a forced yes/no, say plainly what you do not know, and mark the answer as a draft for my review.

Ra-Thor is independent of xAI. Not affiliated with xAI or X Corp.