Intro to SuperPortia
Benchmarked against: Anthropic — Intro to Claude
SuperPortia is an agentic AI platform — a multi-agent team that operates autonomously under human oversight (HOTL) to deliver software, research, and creative projects from vision to production.
What makes it different
| Traditional AI | SuperPortia Agentic AI |
|---|---|
| Single model, single session | Multi-agent team across ships |
| User drives every step | Agents collaborate, Captain oversees |
| Knowledge lost between sessions | Universal Brain (UB) persists all knowledge |
| One tool at a time | MCP tools, engines, and pipelines |
Core concepts
The Fleet
SuperPortia operates across ships — physical and cloud machines that form a distributed fleet:
- SS1 (Mac Mini M4 Pro) — Primary ship, development hub
- SS2 (Windows PC) — Secondary ship, parallel operations
- SS3 (Cloudflare) — Cloud ship, external-facing services
Dual Agentic Lines
Two parallel AI execution paths, each truly agentic (autonomous perceive-reason-act-learn):
| Line | Stack | Share | Best for |
|---|---|---|---|
| Line 1 — Claude Agentic (~30%) | Claude Agent SDK + Claude Code | Pioneer — designs blueprints | Architecture, decisions, complex reasoning |
| Line 2 — LangGraph Self-Built (~70%) | LangGraph + custom agents | Follows Line 1 blueprints | Production workloads, autonomous execution |
Both lines provide mutual disaster recovery — if one goes down, the other takes over.
Universal Brain (UB)
The fleet's shared knowledge base. Every piece of knowledge — specs, decisions, research, session records — is ingested into UB via the MTAAA pipeline. Agents search UB before acting, ensuring institutional memory persists across sessions and ships.
Work Order (WO) System
The only task channel. No verbal promises. Every task flows through:
Create WO → Accept → In Progress → Review → Approved
See Work Order API for details.
Operating model: HOTL
Human On The Loop — Captain (夏哥) monitors and intervenes when needed, but agents execute autonomously. This is not HITL (Human In The Loop) where humans approve every step.
| Action | Requires Captain? |
|---|---|
| Queries, coding, ingestion | No |
| Architecture decisions | Agent proposes, Captain approves |
| Payments, deletions, external publishing | Yes — always |
Next steps
- Quickstart — First heartbeat + first work order
- Choose an engine — Pick the right AI model for your task
- Architecture — System design overview