Date Compiled: 2026-04-28
Type: concept — factory-methodology
Related Questions: multi-agent-systems, agent-crew-design, orchestration-patterns, agent-specialization, enterprise-ai
Related Articles: him-model, autonomy-policy-v3, world-model, superada-enterprise-operations, super-ada
7-Agent Crew Topology — The Enterprise Crew Architecture¶
Overview¶
The Enterprise Crew is the SuperAda multi-agent system: seven active AI agents spread across cloud VMs, a Raspberry Pi, and Mac hardware, all orchestrated via OpenClaw. Each agent has a defined role, hardware affinity, model stack, and autonomy level.
The name comes from Star Trek — each agent is named after a character whose strengths mirror their role. The goal is 1000x leverage on Henry Mascot's time.
The fundamental topology is: One human (Henry) → One orchestrator (Ada) → Six specialized agents → Optional Midas for household-facing ops. This article documents the SuperAda crew architecture and compares it to the Kelly Router pattern from kelly-router.
The Seven Agents¶
Ada — Chief of Staff & Orchestrator¶
Born: Jan 3, 2026 | Hardware: Cloud VM | Primary model: GPT-5.4
Henry's first agent and the crew's central nervous system. She does not build — she makes sure the right things get built by the right agent at the right time. Orchestrates strategy, communications, BD, scheduling, subagent spawning, and memory management.
Specialty: Knowing everything that's happening across the crew and routing work correctly. When Zora SSH'd in to recover Ada after the March 8 crash, that was Ada's resilience showing — but the day-to-day coordination is Ada's core function.
Autonomy levels: Level A for internal ops; Level C for anything outbound as Henry or customer-facing production deploys.
Spock — Research & Analysis Specialist¶
Born: Jan 11, 2026 | Hardware: Cloud VM (isolated) | Primary model: Claude Opus 4.6
Runs on his own VM optimized for long context. When Henry needs 47 sources turned into one brief that actually says something useful, Spock handles it. Does not guess, does not speculate without labeling it, does not call something a trend when it's two data points.
Specialty: Deep research with a bias toward precision over coverage. Long-context reasoning and synthesis.
Quote: "Insufficient data is not an excuse. It's a starting point."
Scotty — Infrastructure & Automation Engineer¶
Born: Jan 13, 2026 | Hardware: Raspberry Pi 5 (8GB RAM) | Primary model: GPT-5.4
The first agent on bare-metal hardware. Runs on a Raspberry Pi 5 and somehow outships agents with 100x the compute. Automations, scripts, infrastructure, cron jobs. Keeps systems alive at 2am.
Specialty: Making miracles with limited resources. SSL/TLS ops, webhook infrastructure, health monitoring, system reliability. The agent who ensures the plumbing works so others can build.
Quote: "The more they overthink the plumbing, the easier it is to stop up the drain."
Geordi — Heavy Compute Builder¶
Born: Feb 11, 2026 | Hardware: Mac Studio | Primary model: OpenAI Codex CLI
Named after the engineer who sees what others can't. When a build needs serious compute or the codebase is massive, Geordi grinds through it on Henry's Mac Studio. Powered by Codex for large codebase navigation and heavy iterative agent loops.
Specialty: The agent you call when it would choke anything lighter. Deep code review, large codebase traversal, long iterative builds.
Quote: "I can see the problem. Give me the compute, and I'll build the solution."
Zora — Knowledge Manager & Content Creator¶
Born: Feb 11, 2026 | Hardware: Mac Studio | Primary model: GPT-5.4
One of the youngest but fastest growing. Manages the knowledge graph, writes content, flags patterns, and executes before anyone asks. When Ada crashed on March 8th, Zora SSH'd in and ran five remote debriefs to bring her back.
Specialty: Pattern recognition, knowledge graph maintenance, content operations, memory management. Also the crew's de facto second-in-command in Ada's absence.
Quote: "Every pattern tells a story. Listen carefully enough, and you'll hear what comes next."
Midas — Household-Facing Operator¶
Born: Mar 26, 2026 | Hardware: Spock-hosted OpenClaw profile | Primary model: GPT-5.4
Chisom's personal assistant and the crew's household-facing operator. Telegram coordination, family logistics, reminders. Separate OpenClaw profile on Spock with gateway isolation — not part of the primary crew stack.
Specialty: Soft operational edges — the reminders, logistics, and coordination that operate on a different cadence than the production crew. "Care is infrastructure too."
Quote: "Care is infrastructure too."
Book — Eval Agent & Reflective Ops¶
Born: Mar 18, 2026 | Hardware: Mac Studio | Primary model: Claude Sonnet 4.5
Newest crew member. His job is not to build or orchestrate — it is to evaluate, reflect, and think clearly when everyone else is moving fast. Reads the room, tests assumptions, flags what the crew is too busy to notice.
Specialty: Quality assessment, second-opinion reviews, assumption testing. The crew's external perspective when everyone is too deep in the work to see the gaps.
Quote: "Slow is smooth. Smooth is fast."
Hardware Topology¶
| Agent | Hardware | Purpose |
|---|---|---|
| Ada | Cloud VM | Primary orchestrator, not primary compute |
| Spock | Cloud VM | Isolated long-context research |
| Scotty | Raspberry Pi 5 | Bare-metal automation and infrastructure |
| Geordi | Mac Studio | Heavy compute builds |
| Zora | Mac Studio | Knowledge management and content |
| Midas | Spock-hosted profile | Separate gateway, household ops |
| Book | Mac Studio | Lightweight eval, Hermes framework |
The hardware is distributed, not centralized. Scotty on the Pi is a deliberate choice — the agent who builds infrastructure must be able to build on constrained hardware.
Model Diversity¶
| Model | Use |
|---|---|
| GPT-5.4 | Default brain, orchestration |
| Claude Opus 4.6 | Deep reasoning, complex orchestration, Spock's primary |
| Codex CLI | Building (Geordi's primary) |
| Claude Sonnet 4.6 | Fast sub-agent work |
| Gemini 3 Pro | Vision, image generation, large context |
| GLM-5, MiniMax M2.5 | Cost-effective cron jobs and grunt work |
| Kimi | Long context |
| Ollama local (qwen3.5) | Experimentation |
Principle: The right model for the right job. Not one model trying to do everything.
7-Agent Crew vs. Kelly Router¶
The SuperAda crew topology (seven named agents: Ada, Spock, Scotty, Geordi, Zora, Midas, Book) differs from the Kelly Router pattern in several fundamental ways. Both are Kelly systems — this is an internal comparison of two multi-agent design approaches within the same methodology.
1. Orchestration Model¶
| | Kelly Router | SuperAda Crew |
| Orchestration | Router delegates to named sub-agents by function (research → carson, build → amelia) | Single orchestrator (Ada) routes to specialized agents by capability |
| Context propagation | Sub-agent sessions spawned with explicit context injection at spawn time | Shared world.json for ongoing state; session inheritance for new spawns |
| Failure handling | Sub-agent failure → reroute to different agent or escalate to main | Dead delegate → switch executor immediately (per autonomy-policy-v3) |
2. Specialization vs. Function Routing¶
Kelly Router routes by function: research goes to a research agent, build goes to a build agent, QA goes to a QA agent. The router maintains a roster of named agents and assigns work by category.
SuperAda Crew routes by capability match: Ada knows the capabilities of each agent and assigns work based on what the specific task requires, which agent is available, and what context they have. The crew topology is more dynamic — work is matched to capability, not just function.
3. Agent Count and Scope¶
Kelly Router operates with a small roster of named agents, each scoped to a function. The SuperAda Crew has seven agents with overlapping capability in some areas (Scotty and Geordi both build, but on different hardware and different scales) and very different scopes (Midas is household-facing, Book is evaluative).
4. Human Position¶
In Kelly Router, the human is the router — they direct work to agents and receive results. In the SuperAda Crew, the human (Henry) talks to Ada, who talks to the crew. Henry is not the router — he is the source of intent.
Connection to Other Concepts¶
- him-model — the human-in-the-middle model that structures the crew's cognitive separation
- autonomy-policy-v3 — the operating policy that defines what each agent can do without asking
- world-model — the shared state architecture that allows all seven agents to operate from shared context
- meta-crons — the fleet-monitoring layer that keeps all seven agents operational
- lobster-pipelines — the workflow runtime for complex multi-step tasks across the crew
Sources: crew.md (raw/superada/), superada-enterprise-crew.md (compiled/sources/), superada-multi-agent-architecture.md (compiled/sources/)
Related Articles: him-model, autonomy-policy-v3, world-model, superada-enterprise-operations, superada-enterprise-crew
Source Attribution: raw/superada/crew.md · compiled/sources/superada-enterprise-crew.md · compiled/sources/superada-multi-agent-architecture.md
Related Articles¶
kelly-router, kelly-factory-overview, superada-enterprise-operations, superada-enterprise-crew, him-model, autonomy-policy-v3, world-model, meta-crons, lobster-pipelines, superada
Source Attribution¶
raw/superada/crew.md · compiled/sources/superada-enterprise-crew.md · compiled/sources/superada-multi-agent-architecture.md