Omnipotent vs. SaaS AI Agents
The difference isn't just “local”. It's about ownership, memory, and agency. See how Omnipotent stacks up against the cloud giants.
| Category | iloomReview workflow | OmnipotentAgentic runtime |
|---|---|---|
| Scope | AI code assistant with review workflow | Full Agentic Runtime — 12 roles, memory, protocolsedge |
| Architecture | CLI tool wrapping Claude Code | Kernel + Signal Bus + Docker substrate with Rust and Deno servicesedge |
| Context Storage | Session-based, per-worktree context | Three memory tiers: Local FS (hot), PostgreSQL/pgvector/Qdrant (cold), Supabase + GitHub federation (via Diplomat)edge |
| Agent Model | Single agent with isolated sessions | 12 specialized roles with defined capabilities, fallback chains, and coordinationedge |
| Memory | Resets between sessions | Persistent across sessions — compounding knowledge via Mnemosyne Protocoledge |
| Self-Correction | Human reviews every output | 9-step orchestration protocol with mandatory checkpoint gatesedge |
| Sovereignty | Depends on Claude Max subscription + cloud | Runs on your hardware. Bring your own API keys. No subscription required.edge |
| Team Scale | Parallel worktrees with branch isolation | Federation Protocol — sovereign/alliance domains with Diplomat git sync |
| Code Review | Built-in review UI with analysis and plansedge | Protocol-driven — swarm_state.json blackboard for status visibility |
| Performance | Interpreted, cloud-dependent, variable latency | Compiled runtime, native speed, local-only communicationedge |
| Getting Started | Requires: Node.js, GitHub CLI, Claude Code, Claude Max (paid), iTerm2 | Requires: Docker, Rust. That's it.edge |
Scope
AI code assistant with review workflow
Full Agentic Runtime — 12 roles, memory, protocols
Architecture
CLI tool wrapping Claude Code
Kernel + Signal Bus + Docker substrate with Rust and Deno services
Context Storage
Session-based, per-worktree context
Three memory tiers: Local FS (hot), PostgreSQL/pgvector/Qdrant (cold), Supabase + GitHub federation (via Diplomat)
Agent Model
Single agent with isolated sessions
12 specialized roles with defined capabilities, fallback chains, and coordination
Memory
Resets between sessions
Persistent across sessions — compounding knowledge via Mnemosyne Protocol
Self-Correction
Human reviews every output
9-step orchestration protocol with mandatory checkpoint gates
Sovereignty
Depends on Claude Max subscription + cloud
Runs on your hardware. Bring your own API keys. No subscription required.
Team Scale
Parallel worktrees with branch isolation
Federation Protocol — sovereign/alliance domains with Diplomat git sync
Code Review
Built-in review UI with analysis and plans
Protocol-driven — swarm_state.json blackboard for status visibility
Performance
Interpreted, cloud-dependent, variable latency
Compiled runtime, native speed, local-only communication
Getting Started
Requires: Node.js, GitHub CLI, Claude Code, Claude Max (paid), iTerm2
Requires: Docker, Rust. That's it.
iloom says
“Nobody wants to review your AI's work. Not even you.”
Omnipotent answers
Your agents don't answer to a SaaS. They answer to you.
The Key Difference
iloom is a better leash.
Omnipotent is a living team.
One gives you guardrails around a single agent. The other gives you a sovereign workforce with memory, self-correction, and 12 distinct specializations — on your own hardware.
The Mana System
When you use a cloud AI agent, you pay a hidden tax: unpredictable API bills, rate limits, and black-box pricing. The Mana System tracks every token, every inference, and every action against your local energy budget.
Omnipotent meters your own resources. You set the budget. You control the cost.