OMOS Architecture • OIPS-aligned

Portable intelligence, not provider lock-in.

OneGodian intelligence is separated into portable capabilities so model providers remain replaceable adapters rather than becoming the architecture itself.

Instructions / Rules

Identity, purpose, behavior, permissions, formatting, and human-authority requirements.

Knowledge / RAG

Versioned, provenance-aware authoritative context retrieved at runtime.

Skills

Reusable task capabilities composed from rules, retrieval, tools, approvals, and expected outputs.

Tools / Actions

Authorized execution through OneGodian APIs, connectors, plugins, and MCP services.

Tests / Verification

Maturity checks, source validation, policy checks, provenance, and audit evidence.

Runtime authority chain

Human authority ↓ ACC — registration, permissions, action governance ↓ O-H-I / OMOS — evidence, synthesis, policy, decision support ↓ OneGodian LLM Synthesis Engine — provider/model coordination ↓ api.OneGodian.org — authorized execution backbone ↓ Plugins / Tools / Registries / External services ↓ Verification / audit / human review

Knowledge rule

Canonical OneGodian knowledge remains under OneGodian control. Provider-native file stores, assistants, plugins, or model-specific prompts are representations and caches—not the sole source of truth.

Execution rule

Models may recommend or request actions, but authoritative execution belongs to scoped tools and services. Sensitive or legally operative actions remain subject to explicit permissions and human approval where required.