Features Overview
Benchmarked against: Anthropic โ Features overview Scope: All SuperPortia capabilities at a glance
SuperPortia is a multi-agent team system where AI agents collaborate to execute tasks from vision to delivery. This page provides an overview of the key subsystems.
Dual agentic linesโ
SuperPortia runs two parallel agentic lines โ not "choose one", but both simultaneously for mutual disaster recovery:
| Line | Engine | Share | Role |
|---|---|---|---|
| Line 1 โ Claude Agentic | Claude Code + Agent SDK | ~30% | Pioneer โ designs architecture, establishes patterns |
| Line 2 โ LangGraph Self-built | LangGraph + custom agents | ~70% | Builder โ implements the blueprint from Line 1 |
Key principle: "Agentic, not just agent" โ these are autonomous goal-driven systems that perceive, reason, act, and learn. Not chatbots with tools.
Mutual rescue: If one line goes down, the other takes over. Line 2 learns from Line 1 and can cherry-pick non-lock-in components.
Line 1 โ Claude Agenticโ
Claude Code as the engine. Give an agent a task and the SDK runs the autonomous loop with built-in file, shell, and browser tools.
- ๅฐๅ (Claude Code CLI) โ Chief Engineer, architecture and decisions
- ๅฐ่ฅฟ (Claude Chat) โ Strategist, reviews and analysis
- Agent SDK โ Python/TypeScript for custom agent behavior
Line 2 โ LangGraph Self-builtโ
Build your own agent loop. You control every node, manage state graphs, and wire your own tools.
- superportia_agent.py โ Custom LangGraph agent implementation
- supervisor.py โ Flat/nested mode auto-switching for different engines
- UBI Router โ Content routing pipeline (MTAAA)
Universal Brain (UB)โ
The company knowledge base โ all knowledge goes here, nowhere else (Constitution ยง1).
| Feature | Technology |
|---|---|
| Storage | Cloudflare D1 (SQLite) |
| Vector search | Cloudflare Vectorize + Gemini embedding-001 (768d) |
| Keyword search | SQLite FTS5 |
| Hybrid search | Semantic + keyword results merged and ranked |
| Classification | MTAAA pipeline โ 3D automated classification |
Key toolsโ
| Tool | Purpose |
|---|---|
ingest_fragment | Add knowledge to UB |
search_brain | Find knowledge in UB |
intel_search | Web search + auto-ingest to UB |
Governanceโ
All UB entries must follow UB Governance rules: English language, controlled vocabulary tags, quality checklist, and freshness management. These rules are enforced by the MTAAA pipeline.
Work Order (WO) systemโ
The only task channel โ no verbal promises, no chat-based task assignments (Constitution ยง3).
Captain creates WO โ Agent accepts โ Agent executes โ Submit for review โ Captain approves
| Feature | Detail |
|---|---|
| ID format | WO-YYYY-MMDD-NNN (auto-generated) |
| State machine | 8 states with enforced transitions |
| RBAC | Captain, PM Agent, Assignee, Requester โ different permissions |
| Transition gates | Some transitions require fields (e.g., completion_summary for review) |
See Work Order API for full API reference and WO Status Transitions for the state machine.
MCP Toolsโ
SuperPortia agents interact through MCP (Model Context Protocol) tools โ standardized interfaces that work across all agents and ships.
Tool categoriesโ
| Category | Tools | Purpose |
|---|---|---|
| Knowledge | search_brain, ingest_fragment, get_entry, get_recent | UB read/write |
| Intelligence | intel_search, search_web, run_patrol | External research |
| Execution | dispatch_work_order, call_model | Task dispatch to engines |
| Work orders | create_work_order, accept_work_order, complete_work_order | WO lifecycle |
| Communication | send_agent_message, check_agent_mailbox | Agent-to-agent messaging |
| System | agent_heartbeat, factory_floor_status, sre_status | Fleet monitoring |
| File ops | ubi_read_file, ubi_write_file, ubi_edit_file, ubi_run_command | Local file operations |
See MCP Tools for complete tool documentation.
Engine fleetโ
Multiple AI engines for different cost/capability tradeoffs:
| Engine | Cost | Best for |
|---|---|---|
| Claude (Opus/Sonnet/Haiku) | $$$$ | Architecture, decisions, code ops |
| Gemini (Flash/embedding) | $ | Search with citations, embeddings |
| Groq (Llama) | Free | Quick research, trivial tasks |
| DeepSeek | ยข | Analysis, reasoning |
| Mistral | ยข | European model, alternative |
| Zhipu (GLM-5) | ยข | Chinese NLP, agent tool-calling |
Selection principle: CP value is NOT "cheapest possible" but "minimum cost that gets the job done RIGHT" (Engine Selection Guide).
Fleet architectureโ
Three ships running the same agent stack:
| Ship | Platform | Location | Status |
|---|---|---|---|
| SS1 | Mac Mini M4 Pro | ๅฐๅๅทฅไฝๅฎค | ๐ข Running |
| SS2 | Windows PC | ๅฎถไธญ | ๐ก Building |
| SS3 | Cloudflare (cloud) | Global edge | ๐ด Planning |
All ships share the same Cloud UB, same governance rules, same WO system. The fleet is designed for mutual monitoring โ if one ship goes down, others detect and compensate.
Governance frameworkโ
| Document | What it governs | Link |
|---|---|---|
| EGS | Engineering rulebook (9 chapters) | EGS Spec |
| Company Constitution | 14 supreme articles all agents must follow | Constitution |
| UB Governance | Knowledge ingestion rules | UB Governance |
| Agent Intelligence Protocol | 8 mandatory agent behaviors | AIP |
Related pagesโ
| Page | Relationship |
|---|---|
| Work Order API | The task system API |
| WO Status Transitions | State machine rules |
| MCP Tools Overview | All available tools |
| Engines Overview | Engine fleet details |