Happy slice of mcp pizzaMCP.pizza

arifOS

MCP.Pizza Chef: ariffazil

arifOS is a server that acts as a constitutional judge for AI agents. It evaluates AI actions against 13 strict ethical and safety rules before allowing them to proceed. Instead of executing commands, it returns clear verdicts like approved, hold for review, or blocked. It works with any AI system using the MCP protocol and requires no complex setup, making AI actions safer and auditable in real time.

Other

Use This MCP server To

Check if an AI action is safe to execute Get a clear approval or hold verdict for AI decisions Enforce ethical rules on AI behavior Audit AI decisions with immutable logs Pause AI actions needing human review Ensure AI respects human vetoes Track AI action history securely

README

arifOS

A constitutional control plane for AI agents.

arifOS evaluates identity, evidence, authority, risk, and reversibility before an AI agent may execute a consequential action.

It does not execute actions itself. It returns an auditable verdict:

  • SEAL — authorized under stated conditions
  • HOLD — insufficient evidence or human approval required
  • SABAR — proceed cautiously, partial authorization
  • VOID — blocked by a constitutional floor

Try the safe demo · Connect through MCP · Read the constitution · Documentation

The agent that acts is not allowed to certify its own action.

DITEMPA BUKAN DIBERI. Forged, Not Given.

What arifOS does

arifOS is a Model Context Protocol server that sits between an AI agent's intent and the action it wants to take. It enforces 13 constitutional floors — non-negotiable rules about truth, reversibility, humility, auditability, and human authority.

When an agent calls arif_init, arifOS binds a session, activates the floors, and routes the agent through a governed loop: observe evidence, think under constraints, dispatch to the right organ, recall governed memory, receive a constitutional verdict, execute (if authorized), and seal the result.

When an agent's evidence is thin, risk is high, or an irreversible action is proposed — arifOS returns HOLD. A HOLD is not a failure. It is a fence.

What arifOS does not do:

  • It does not execute actions (A-FORGE executes; arifOS judges).
  • It does not wrap an LLM (it governs any agent that speaks MCP).
  • It does not replace human judgment (F13 = human veto is final).

The 8 canonical verbs

arifOS exposes 8 verbs on its public MCP surface. Each verb corresponds to a station in the 000–999 authority ladder.

Station Verb What it does
000 arif_init Bind session, activate floors, mint token
111 arif_observe Sense reality — search, fetch, vitals, entropy
333 arif_think Structured reasoning under F2/F7 constraints
444 arif_route Dispatch intent to the correct organ
555 arif_memory Governed recall, store, revise (L1–L6)
666 arif_judge Constitutional verdict: SEAL / HOLD / SABAR / VOID
777 arif_forge Governed execution (post-SEAL, lease-gated)
999 arif_seal Immutable append to VAULT999 ledger

Canonical stage assignments live in tools_sot.yaml, sourced from constitutional_map.py. If prose and code disagree, code wins.

The 13 constitutional floors (F1–F13)

These are checked on every governed action. They are the physics of the kernel.

arifOS FAQ

Can I use arifOS to ensure AI actions follow strict ethical rules?
Yes — arifOS judges AI actions against 13 constitutional rules before approval.
Can I use arifOS to get a safety verdict before executing AI commands?
Yes — it returns clear verdicts like SEAL (approved) or HOLD (needs review).
Which apps or clients work with arifOS?
arifOS works with any AI agent or client speaking the MCP protocol; it is model-agnostic and can be used with GPT-4, Claude, Gemini, etc.
Do I need an API key or account to use arifOS?
No special API key is needed; it is accessible via a remote URL for easy setup.
How hard is it to set up arifOS?
Setup is one-click via a remote URL, no complex installation required.