# Install Wizard — Build Your Own Nexus-Style Executive Team
### Hand this to your AI assistant. It runs the intake, then the build sequence.

> Goal: replicate the pattern from Brendan Hsu's build session — one harness, an executive team
> of specialist agents, multiplayer in the tool your team already uses. Model-agnostic, API-first,
> so nothing is locked to one vendor.

## Prerequisites (the wizard confirms these first)
- A comms surface your team already lives in (**Slack** recommended; Telegram/WhatsApp work).
- A harness/orchestrator (Brendan uses **Hermes**) that can hold agents + tools + memory.
- A default model via **OAuth or API** (ChatGPT/Claude) + **OpenRouter** as a swap-in fallback.
- A memory layer (start simple: one store; mem0/Supabase/Obsidian/Pinecone are options — don't
  over-build; AI Hub's house pattern is a single namespaced D1/Postgres table).
- Read access to what you already have: Google Drive, docs, calendars, your PM tool, Stripe.

## Intake questions (the wizard asks the member)
1. What does your team use for daily comms today? (that's your "steering wheel")
2. List your top 5 tools/systems (PM, CRM, storage, billing, comms).
3. What are the 3 recurring deliverables you wish an assistant could produce? (proposals?
   sites? financial models? CRMs?)
4. Who are the "functions" you'd staff as agents? (CEO, sales, CFO, marketing, design, research…)
5. For each function, who are the thought leaders whose approach the agent should ingest?
6. What's the ONE workflow that, if automated, saves the most hours this week? (build that first)

## Build sequence
1. **Lay the plumbing.** Connect comms + storage + PM + billing to the harness via OAuth/API.
   Confirm the agent can read your existing docs and act, not just chat.
2. **Stand up the primary driver** (Brendan's = Atlas). One router/brain, reachable in Slack.
3. **Add specialist personas** — one per function. Instruct each to ingest its domain's thought
   leaders and adopt their structure/voice. (Don't spin up 13 separate bots — one brain, many
   personas, scoped data per channel.)
4. **Go multiplayer.** Give each team member a channel; drop the relevant agents in; teach by
   *doing* in real time — this is the fastest adoption path.
5. **Ship the first deliverable** from intake Q6. Prompt + screenshot of what "good" looks like;
   iterate where it breaks. Taste is the differentiator — feed brand kit, style guide, voice.
6. **Wire proactivity.** Connect meeting capture (Fireflies) + cron so agents have context
   before every meeting and can produce proposals/roadmaps unprompted.

## Guardrails / learnings from the session
- **Model-agnostic or bust.** Build API-first so you can swap the engine (chassis vs. engine).
- **Keep memory lean.** Don't copy a 4-system memory sprawl; start with one store.
- **Taste matters.** "If it looks like ass, I wouldn't take you seriously" — hire a designer
  for a few hundred bucks if needed; feed the system a real brand kit.
- **Security:** run a prompt-injection test on any client-facing voice/chat agent before ship.
- **The operator is the unlock.** The system amplifies someone who already knows what they want.

## NEEDS HUMAN INPUT
Exact harness install steps, credentials/OAuth scopes, cost, and data-privacy protocols are
environment-specific — confirm with your implementation partner before publishing to a team.