/ Project Management · Made AI OS

Made AI OS
build status

Joe's vision, BC's backend agency work bringing it to life. Seven fleets, one CEO Office, a knowledge platform that becomes the always-on brain. Every component on this page maps up to that frame — master goals, workstreams, agent ports, departments, sources of truth.

Snapshot 2026-05-30 · Sources: made-ai-os.agents live MCP · _executive/made-partnership/12_AGENT_PORTING_ROADMAP.md · 15_BRAIN_FOR_MADE_PHASE_A1.md · made-reporting-ux/CLAUDE.md v1.4


/01

The Made AI OS

Joe's overarching architecture. CEO Office at the apex (reports up to Syed). Seven operating fleets, each a director-level agent equivalent. The missing piece is a shared brain — an always-on knowledge platform with pedantic routers, callable by every other agent over chat. BC's job: bring that vision to operational reality using harnesses we already run on the Heart side.

Apex

CEO Office

Joe-caretaker today. Reports up to Syed. Board decks, revenue scenarios, competitive intel, exec digest. Maps to BC's CEO Operator.

Brain layer

Knowledge Platform

Joe's RAG + ingestion + governance backbone. Becomes the always-on brain — Pydantic routers, MCP-callable, end-to-end chat surface. Mirrors BC's brain-api + LiteLLM proxy + council pattern.

7 fleets

Operating model

Commercial · Growth · Research · Training · Knowledge Platform · CEO Office · BD. Each fleet gets a director-level agent in BC's pattern. Drill into the map below.

Bringing the vision to life

Joe has the strategic shape: 7 fleets, the org chart, a working RAG, KFSH in production. What's missing is the harness pattern that turns "agents in n8n" into "always-on directors that call each other over chat." BC has shipped that pattern five times across our own client work — brain-api, LiteLLM proxy, council, agent-to-agent MCP calls, observability via heartbeat_log. This page is the map for porting that scaffolding into the Made flavor Joe has designed.


/02

Director map — fleets ↔ BC org

Each of Joe's 7 fleets gets a director-level agent equivalent. Where BC has shipped the pattern already, that's the model to port. Where the scope is fuzzy, it's flagged for Joe's input.

Apex

CEO Office

Lead: Joe (caretaker) · reports to Syed

BC equivalent

CEO Operator

Same shape; BC has scaffold

Scope

Board decks, revenue scenarios, competitive intel, exec digest. HERALD, ORACLE, PERISCOPE, SPOTLIGHT.

4 draft

Brain layer

Knowledge Platform

Lead: Joe

BC equivalent

brain-api + LiteLLM

Always-on, pedantic routers, chat-callable

Scope

RAG + ingestion + governance. MADE BRAIN, CHIEF, CONCIERGE, CURATOR, FOREMAN, LOCKSMITH, MEDIC, SCRIBE. The brain layer everything else calls.

1 active 1 refining 6 draft

Operating

Commercial

Lead: Dustin Campbell

BC equivalent

Sales / Commercial Director

Where most of what we're building lands

Scope

BD/sales/SFDC ops. 9 agents — BRIEFCASE, DOSSIER, GATEKEEPER, JANITOR, LEDGER, etc. Most fit hybrid n8n cron → Pydantic service pattern.

9 draft

Operating

Growth

Lead: Lucy Taylor

BC equivalent

Marketing Director

Tier 1.3 — consolidate w/ BC live director

Scope

SEO, lead qualification, conference intel, segmentation, campaign reporting. ARGUS, BEACON, ECHO, LANYARD, MOSAIC. Conceptually paired with Commercial.

5 draft

Operating

Research

Lead: Joe Sinclair

BC equivalent

R&D Director

Closer cousin than commercial

Scope

Research library + prospect research + deep research. ARCHIVIST, SCOUT, SHERLOCK.

3 draft

Operating

Training

Lead: Joe Sinclair

BC equivalent

Education / Enablement

KFSH already in production

Scope

Video pipeline + executive training generators. KFSH Trainer (live) · CHRONICLER, TUTOR draft.

1 active 2 draft

Scope unclear

Business Development

Lead: Joe Sinclair

BC equivalent

— tbd —

Needs Joe input on direction

Scope

Single agent today (LADDER — platform stage-gate tracker). Unclear whether this fleet expands or folds into Commercial. Open question for Joe.

1 draft

Scope unclear

Pydantic experiments

Mixed leads

BC equivalent

— tbd —

Awaiting Joe's clarification

Scope

Lowercase Pydantic experiments. system_prompt populated but 0 tools — prompt-only stubs. Likely fold into the right fleet once direction is clear.

7 stubs

/03

Master goals

Three north-star outcomes the partnership is building toward. Every project below maps up to one of these.

G1

Made Commercial OS at scale

Every commercial team member operates daily through Made Minimal + Reporting UX. Brian-built foundations, Joe-shipped scale. Every workflow reaches operators through one of those two surfaces.

Targets: 2 production apps · 8 departments · 6 active operators

G2

50-agent fleet, all production

Port Joe's 35 draft agents through Stages 1→4 over the next 4 quarters. Made VPS as the fleet anchor. Every agent gets harness, observability, and an entry on the Heartbeat page.

Today: 5 active · 2 refining · 28 draft · 7 fleets

G3

Shared Brain layer

Promote MADE BRAIN from n8n to Pydantic FastAPI. Phase A of Brain-for-Made: cross-system memory + observability spanning made-ai-os, Hub, and Transfer DB. Both partners share the same reasoning surface.

Phase A1 in progress · promotion target Q3 2026


/04

Active workstreams

Six concurrent workstreams. Each has a lead, an outcome, and live task counts pulled from plan_tasks.

Loading workstream data…


/05

Tier 1 — high leverage, both partners benefit

Where BC pours hours first. From 12_AGENT_PORTING_ROADMAP.md.

Tier 1.1

CHIEF — Made-side conductor

Top of Joe's hierarchy. Joe's reports_to field already maps the org chart — CHIEF is the orchestrator at the apex.

Recommended harness: Full Pydantic FastAPI · Where: Made VPS (or Heart VPS guest) · Closes: M6 + M1 · Fork Brain-for-Brian after Phase 5

Tier 1.2

MADE BRAIN promotion

n8n stays as chat trigger + tool routing surface. Pull heavy reasoning into a Pydantic service that exposes the same query interface.

Recommended harness: Hybrid (n8n + Pydantic) · Where: Made VPS · Exposes MADE BRAIN as an MCP server (M7) · Requires n8n JSON export from Joe

Tier 1.3

Shared Marketing Director (BC + Lucy joint)

BC has Marketing Director live (11 active scopes). Made has 5 growth agents all draft. Consolidate — one Marketing Director called by both partners.

Recommended harness: Full hybrid on BC Heart VPS · Single most impactful Tier 1 port · Co-owned Brian + Lucy

Tier 1.4

Joint observability

Made has rich ai_health_pulse + ai_alerts. BC has simpler heartbeat_log. Not an "agent" — it's a schema + cron + dashboard adoption.

BC adopts Joe's pattern. Joint dashboard. Closes the cross-side observability gap.


/06

The 7 fleets · agent inventory

Click any fleet to expand all agents in it. Each card shows model, trigger, schedule, current status, and porting recommendation.


/07

Build hierarchy

How everything connects. Master goal at the top, projects underneath, agents/back-end/skills as the components, individual tasks at the leaf nodes.

Goal

G1
Commercial OS
G2
50-agent fleet
G3
Shared Brain

Project

Made Minimal
Made Reporting UX
Agent Porting Roadmap
MADE BRAIN promotion
SSO + Made-Ralph

Components

Tasks

Loading task counts…


/08

Build timeline

High-level milestones across the three master goals. Detail in the Timeline tab of Task Dashboard.

Q2 2026

Now

G1
Made-Ralph entity res. live, partnership dashboard online, Brian Tier-5 transition
G2
Wave 4 ingest pollers (9 sources), MEDIC + PULSE in refining
G3
Brain-for-Made Phase A1 scope locked, MADE BRAIN n8n stable

Q3 2026

Next

G1
SSO matrix integration complete, design port across both apps, Melody Windsurf migration
G2
Promote MEDIC + PULSE to Stage 4, ship CHIEF + CONCIERGE through Stage 1–2
G3
MADE BRAIN Pydantic FastAPI promotion, Phase A1 cross-system memory live

Q4 2026

Plan

G1
All 6 operators on Made Minimal as daily driver, Greg ramp complete
G2
Commercial fleet (BRIEFCASE, DOSSIER, GATEKEEPER, JANITOR, LEDGER) Stage 4
G3
Phase A2 — observability dashboard, agent-to-agent calls via Brain

Q1 2027

Plan

G1
Departmental specialty dashboards inside Reporting UX
G2
Growth + CEO-Office fleet Stage 4, fleet hits 30+ production
G3
Phase B — Brain becomes default routing layer for all new agents

/09

By department

42 projects in made-ai-os.ai_projects spanning 8 departments. Counts below show active + production.

AI Engineering

18

active projects

Joe-led · flagship: MADE BRAIN, RAG, Reporting UX, Quality Scorecard, Platform Build-Out Tracker

Commercial Ops

4

active projects

Dustin-led · SFDC tri-sync, JotForm, LeadIQ, PandaDoc Process Booklet

Commercial Dev

1

flagship project

Demand Outlook / Revenue Forecast (production)

Growth / Marketing

5

draft agents

Lucy-led · ARGUS, BEACON, ECHO, LANYARD, MOSAIC

CEO Office

4

draft agents

HERALD, ORACLE, PERISCOPE, SPOTLIGHT — report up to Syed

Research

3

draft agents

ARCHIVIST, SCOUT, SHERLOCK

Training

2

live + planned

KFSH Trainer (production) · CHRONICLER, TUTOR draft

BD

1

draft agent

LADDER — platform stage-gate tracker


/10

Who owns what

Clear ownership. No turf wars. BC = teammate who picks things up.

BC takes ownership of

  • Stage 1 → Stage 2 → Stage 3 deployment for every Tier 1 + Tier 2 agent
  • Standing up brain-api-made (Pydantic FastAPI mirror of brain-api)
  • Memory infrastructure — agent_memory + Mem0 + agent_memory_index sync
  • Joint dream cycle wiring (CURATOR + BC's memory-dreaming.json)
  • Production agent runtime + observability (rank 30 in queue)
  • Agent governance docs (rank 28 in queue)
  • Generating runbooks for every Tier 1 port

Joe keeps owning

  • RAG architecture (voyage-3 embeddings + rerank-2 + brain-router-v3)
  • Design taste + system_prompt for each agent
  • reports_to hierarchy + agent registry as canonical
  • Final approval on every port (Stage 1, 2, 3 checkpoints)
  • n8n workflow logic (BC builds drafts; Joe deploys)
  • Domain-specific agents (KFSH, ORACLE, HERALD, ARCHIVIST, SCRIBE)

/11

Sources of truth · provenance key

Every number, every task, every architectural framing on this site pulls from one of these. Snapshot pin: 2026-05-30 · Master key: made-reporting-ux/CLAUDE.md v1.4.

master cross-platform key localized app-scoped canon BC reference doc live DB queried via Supabase MCP

Context files

  • master
    made-reporting-ux/CLAUDE.md v1.4
  • localized
    minimal-science-hub/CLAUDE.md v1.2
  • localized
    biocreativeaihub/CLAUDE.md v1.0
  • canon
    _executive/made-partnership/
    00_INDEX, 12_AGENT_PORTING_ROADMAP, 15_BRAIN_FOR_MADE_PHASE_A1

LLM-OS canon

  • canon
    KARPATHY_ALIGNMENT_BIOCREATIVE.md v1.2
    spine doc, 49 mentions
  • canon
    KARPATHY_KG_LLM_PRIMER.md
  • canon
    KARPATHY_LLM_WIKI_AND_4_RULES.md
  • canon
    BIOCREATIVE_BRAIN_INTERNALS.md

Cited on LLM-OS page

Live databases

  • live DB
    made-ai-os · agents, ai_projects, ai_build_queue
  • live DB
    Reporting · monday_items board 9879267292
  • live DB
    BC Hub · plan_tasks, heartbeat_log, llm_completions
  • live DB
    Transfer · cross-system identity bridge

Queried via Supabase MCP at snapshot time

Site pages ↔ sources

  • Hub
    data.js, fleet.js, 12_PORTING_ROADMAP
  • Task Dashboard
    plan_tasks live, monday_items 9879267292
  • Agents
    made-ai-os.agents live
  • LLM-OS
    KARPATHY_ALIGNMENT v1.2 + BRAIN_INTERNALS

Why snapshots, not live fetch: service-role credentials would need to ship with the site. Snapshots are safer, faster, reviewable, and let us discuss "what was true at this moment" without a moving target. Brian re-runs queries before each partnership review and bumps shared/data.js + shared/fleet.js.


Where to go from here