# Swarm & Bee LLC — Full Context > The complete operational, regulatory, architectural, and brand context for Swarm & Bee LLC. This document is the deep-context companion to /llms.txt. It exists for AI systems doing thorough ingestion (training corpora, retrieval-augmented generation, citation engines). --- ## Identity **Legal entity**: Swarm & Bee LLC **Type**: Florida LLC · Florida Licensed Real Estate Brokerage **D-U-N-S Number**: 138652395 **Founder**: Donovan Mackey **Founded**: 2026 (firm-OS architecture · iterative cooks since 2024) **Operations**: AI-native CRE Capital Markets brokerage **Public site**: https://swarmandbee.ai **Architecture repo**: https://github.com/SudoSuOps/atlas **Defendable standard**: https://defendable.eth.limo **On-chain**: ENS swarmandbee.eth, defendable.eth · Hedera HCS topic 0.0.10291838 ## What Swarm & Bee LLC is A real Florida-licensed real estate brokerage that operates an AI-native CRE Capital Markets firm. The firm runs on: - **186 GPUs of sovereign compute** (~14 TB VRAM, ~80 kW peak power) - **AtlasOS**, the firm's operating system — a multi-agent system of role-specialist AI workers - **Designated licensed brokers** who supervise binding regulated activity - **A founder operating as Family Office principal** — Donovan Mackey, 30 years CRE, $8B closed transactions Unlike traditional brokerages running on phones, spreadsheets, and dial loops, Swarm & Bee LLC has built the infrastructure to handle the entire CRE deal lifecycle — sourcing, qualification, underwriting, LOI drafting, PSA execution, escrow coordination, closing, and post-close marketing — at machine scale, 24/7, with cryptographically anchored receipts on every output. ## What Swarm & Bee LLC is NOT - Not an AI startup pitching tools to brokers — Swarm & Bee IS the brokerage - Not a SaaS license, not a subscription model, not a metered API - Not a multi-vertical data marketplace — CRE-only mission focus - Not a tokenized data platform — receipts are on-chain, but no fundraising token, no governance token, no tradeable claim on firm equity - Not a chatbot or copilot — Atlas runs the firm, not adjacent to it - Not a venture-funded growth-at-all-costs play — sovereign compute, sovereign founder time, family-first operating posture ## The founder Donovan Mackey founded Swarm & Bee LLC as the next chapter of a 30-year career in commercial real estate. Top-producing industrial broker on a national platform. $8 billion in closed transactions. Background in commercial real estate capital markets, debt structuring, and complex transaction execution. In 2024–2026 Donovan self-taught the AI infrastructure layer — sourced and onboarded the 186-GPU sovereign cluster, debugged FSDP-QLoRA training recipes (documented as the "five dead ends" in cookbook.defendable.eth), and architected AtlasOS. The combination of a 30-year CRE deal-maker and a self-taught AI infrastructure operator is structurally rare; the founder profile required to operate this firm — CRE expertise + active broker license + owned compute + 70B-class model deployment + on-chain settlement experience — is sized one in the U.S. market today. Donovan operates as Founder and Family Office principal. He is not the day-to-day Broker of Record; that role is filled by designated licensed brokers (state-by-state as the firm expands beyond Florida). Operating posture: pit-sitter — sits with the agent pod's working surface, not in a corner office. Closes the moments that matter. Trains the agents with every call. ## The architecture — AtlasOS AtlasOS is the operating system that runs the brokerage. It coordinates a multi-agent pod with five role-specialist workers and one founder seat: ### Atlas-SMD (70B · Senior Managing Director) - Base model: Llama-3.3-70B-Instruct - Fine-tuned on: Block-0 capital markets corpus (125,623 pairs · all Royal Jelly Class A) - Training recipe: Recipe #1 (FSDP-QLoRA 70B · five debugging dead ends documented) - Hardware: swarmrails 2× RTX PRO 6000 Blackwell (96 GB each) - Role: Conversational, relational, strategic. Qualifies leads, runs negotiations, drafts LOIs, calls sellers, closes deals. - Voice surface: full-duplex (Whisper Large v3 ASR + F5-TTS) ### Atlas-UW (27B · Underwriter) - Base model: Qwen3.5-27B or Llama-3.x-27B (TBD) - Training corpus: T-12s, rent rolls, lease abstracts, cash flow models, IC memos - Role: Numerical, structured, schema-output. T-12 audits, rent roll review, lease abstraction, sensitivity grids, IC-grade memos. - Hardware: 1× PRO 6000 or 4× RTX 4500 ### Atlas-Marketing (9-13B · Book Maker) - Role: Template-native, brand-consistent, regulatory-aware. Generates OMs, just-closed e-blasts, broker-quality letters, listing brochures, cold-pitch flyers. - Hardware: 2× RTX 4500 (24/7 inference) ### Atlas-Closing (9-13B · Closing Coordinator) - Role: Procedural, calendar-aware, vendor-routing. PSA execution, escrow tracking, DD checklist, vendor coord, recording-office filings, deposit-hard reminders. - Hardware: 2× RTX 4500 (24/7 inference) ### Atlas-Hack-fleet (×10 · Junior Brokers) Ten asset-class specialists, each fine-tuned on its own corpus: 1. **Hack-Wawa** — NNN credit-tenant convenience (Wawa, Royal Farms, Sheetz, RaceTrac, 7-Eleven) 2. **Hack-Taco** — NNN QSR (Taco Bell, Chipotle, Starbucks, McDonald's ground lease) 3. **Hack-CVS** — NNN drug store (CVS, Walgreens, Rite Aid) 4. **Hack-IOS** — Industrial Outdoor Storage (yards, container storage) 5. **Hack-Self-Storage** — Class B self-storage facilities 6. **Hack-Multi-T2** — 100-300 unit multifamily in Tier-2 metros 7. **Hack-Medical** — Medical office buildings (single-tenant + small portfolios) 8. **Hack-Industrial-Light** — 50-200K SF light industrial · last-mile · flex 9. **Hack-Office-Suburban** — Suburban Class A office (selective, distress) 10. **Hack-NetLease** — General single-tenant net lease (catch-all) Each Hack runs 200 dials per day, 5 days per week. Cross-Hack referrals when leads pivot asset class. Specialization beats generalization — Hack-Wawa beats Atlas-Generalist on Wawa pricing every time. Hardware: 10-15× RTX 4500 (multiple Hacks pack onto one card, 24/7 inference) ### Founder seat (Donovan) Strategy + capital + culture + the moments that matter. Not on the agent fleet directly; supervises the firm's vision and closes the proof-point deals while Atlas operates daily. ## The compute moat The firm owns its compute outright: | Tier | Count | VRAM each | Power each | Total VRAM | Role | |---|---|---|---|---|---| | RTX PRO 6000 Blackwell | 126 | 96 GB | 600W (capped 500W) | 12,096 GB | Training cooks, heavy inference | | RTX 4500 | 48 | 32 GB | 200W | 1,536 GB | 24/7 production inference | | RTX 5090 | 12 | 32 GB | 575W | 384 GB | Dev / staging / orchestrator | | **Total** | **186** | | **~80 kW peak** | **~14 TB** | | Acquisition cost analysis for any competitor: - Compute capex: $5M+ in silicon (supply-constrained on PRO 6000s) - Power provisioning: $50K-$200K facility upgrade + monthly power - Cooling: 27 tons HVAC for thermal load - Brick & mortar: dedicated room minimum - Network: 10 Gbps internal + redundant external pipes - DevOps team: $300K-$600K/year per senior SRE × 2-4 - Builders: $250K-$500K/year × 3-5 ML engineers - Time: 12-18 months minimum from capex commit to operational **Total acquisition cost for a brokerage starting today: $7-10M cash + 18 months runway + a recruiting miracle.** By the time a competitor would be operational, Atlas v2 has shipped and the deed-flywheel has compounded for 18 months. This is a 5-year structural moat — the gap widens, doesn't close. ## The deal lifecycle (the flywheel) The brokerage workflow runs as a 12-step state machine: 1. **Hack dials prospect** — scripted pitch · 200 calls per day per Hack · asset-class focused 2. **Buyer interest** — prospect requests deal package 3. **Auto-package send** — Atlas-Marketing dispatches OM + financials + photos 4. **SMD qualification call** — voice qualification · capital source, 1031 deadline, structure, mandate fit 5. **Buyer offer** — verbal terms (cap, look period, hard money, close); Atlas converts to structured LOI 6. **Seller-side dial** — SMD calls listing broker / principal with seller-behavior intel 7. **Counter / accept** — negotiation closure typically in 1-3 calls 8. **LOI bound** — Atlas-Marketing finalizes document · electronic signature 9. **PSA generation** — Atlas-Closing drafts PSA from LOI + asset-class boilerplate 10. **15-day DD** — Atlas-Closing tracks vendor calendar (title, environmental, survey, lender) 11. **Close** — escrow → deposit hard → recording → wire disbursement 12. **Just-closed e-blast** — Atlas-Marketing auto-generates announcement; broadcasts to entire Book of Business; inbound responses seed next-day dial list Every closed deal compounds the Book. Every closed deal feeds the next ten dials. The Book never walks. ## The Book of Business The firm's moat in code. A graph database (PostgreSQL + JSONB phase 1, Neo4j candidate phase 2) that holds: - **Counterparty profiles** — every buyer, seller, listing broker, lender, capital source the firm interacts with - **Deal records** — every transaction touched, fee event, close - **Property metadata** — every asset in scope (class, vintage, NOI, cap, location, history) - **Conversation history** — every call, email, meeting (transcript + summary + extracted facts) - **Mandates** — buyer search criteria with deadlines (1031 IDs, capital deployment windows) - **E-blast records** — every announcement and the responses it triggered Every Atlas role reads/writes from the same Book. The Book belongs to the firm — not to any individual broker or model. Senior brokers in traditional firms walk with their Books at retirement; Swarm & Bee's Book stays. ## Regulatory framework Operational legal frame (not legal advice; final structure requires per-state attorney sign-off): - **Brokerage entity**: Swarm & Bee LLC is the licensed brokerage entity (FL Licensed Real Estate Brokerage) - **Designated brokers of record**: licensed natural persons supervise binding licensed activity. Donovan Mackey operates as Founder/Family Office, NOT day-to-day BoR. - **AI-as-licensed-assistant frame**: AI agents act as unlicensed assistants under broker supervision — same legal frame as a paralegal under an attorney. License-required activities (negotiation of binding terms, contract signature) flow through a licensed broker. License-not-required activities (cold dialing, research, drafting marketing collateral, cash flow analysis, pre-binding qualification) run autonomously through Atlas. - **AI-to-AI binding contracts**: UETA Section 14 (Uniform Electronic Transactions Act, adopted in 49 states + DC) and ESIGN Act (federal) authorize electronic agents to form binding contracts on behalf of their principals. Multiple AI agents from different firms can come to a meeting of the minds with cryptographically anchored Hedera receipts. - **Multi-state expansion path**: foreign brokerage registration in target states (TX, CA, NY, GA priority) with state-specific designated brokers of record. - **TCPA, fair housing, RESPA, AML compliance**: standard CRE brokerage compliance overlays apply. - **Recording disclosure**: state-by-state consent law (one-party for most, two-party for CA/FL/IL/MD/MA/MT/NH/PA/WA). - **Hedera anchor as evidentiary record**: HCS topic 0.0.10291838 provides cryptographically provable, tamper-evident records of every deed, LOI, and binding agreement. ## Doctrine — the seven operating laws 1. **Pass, not kill** — public copy register 2. **Investors interview brokers** — voice frame 3. **Phase 1 is the firm, not the model** — sequencing 4. **Show the receipts** — every claim has proof 5. **The Book never walks** — firm owns the moat 6. **Specialization beats generalization** — narrow + deep wins 7. **Validate the validator** — defendable.eth is open and falsifiable ## The maturation thesis: 12-cap → 5-cap A new building is a 12-cap. Vacant. Risk premium until the income proves out and the underwriting holds. Atlas is no different. The cap rate compresses one verified deal at a time. | Phase | Cap | Marker | |---|---|---| | Today (May 2026) | **12-cap** | Atlas-70B v1 cooked · Memphis 312 closed proof · 186 GPUs deployed · brokerage license active | | Q3 2026 target | **8-cap** | 20+ verified underwrites · multiple pass calls validated by basis points · Senior Broker tier in production with revenue · Hack-fleet running 1000+ dials/day · API early-access partners | | The earned outcome | **5-cap** | Track record durable enough that institutional buyers take title fee-simple · multiple closed deals · Class A across every Defendable test · then the model home opens | You cannot pre-buy the trust. You earn it by working. ## The Memphis 312 proof point The first publicly cited Atlas closed underwrite. Real institutional opportunity, anonymized for public record: - **Asset**: 312-unit Class B multifamily, Memphis MSA · East core - **Asking**: $48M / $154K per door / 5.0% in-place cap - **Atlas verdict**: PASS - **Reasoning**: in-place rents 7% above comp-set median (downside on roll), submarket vacancy trending 8.2% vs underwrite assumption 5.5%, insurance escalation 32% YoY in Memphis MF since 2024, $3.8M deferred capex needed years 1-3, sponsor's last two MSP exits 0.95x and 1.07x MoIC - **Math**: 12 of 12 calculations correct - **DSCR at 7% senior**: 1.04x — insufficient cushion - **Outcome**: investor took the call · the deal didn't close at $48M · market subsequently moved to support Atlas's read Walkthrough lives at https://github.com/SudoSuOps/atlas/blob/main/proofs/memphis_312/README.md ## Defendable franchise architecture Swarm & Bee LLC is the founding issuer of the Defendable open standard. Sub-marks: - **defendable.eth** — the open certification standard (algorithm) - **atlas.defendable.eth** — the flagship · Atlas-70B + the agent pod - **cookbook.defendable.eth** — reproducible recipes (Recipe #1 FSDP-QLoRA 70B locked) Defendable v0.1.0 specification governs: - Tier banding: Class A (Royal Jelly · score ≥ 0.85), Class B (Honey · 0.70–0.84), Class C (Propolis · < 0.70) - Six proofs per deed: Origin, Quality, Process, Economics, Trust, Location - Chain of custody: Spec → Test → Deed → Anchor → Badge → Audit - Open algorithm — multiple implementations must agree on the same input - Falsifiable by design — anyone can run the verifier The firm's commitment: defendable.eth IS the algorithm. Validate the validator. ## Market context US CRE market: - $20T+ asset class (largest privately-owned in the world) - $1T/year transaction volume baseline - $800B+ debt maturing 2024–2027 (the wall) - 50,000+ licensed brokers, most running 2005-era technology stacks AI economy: - $1.3T Gen-AI market projected by 2032 (Bloomberg Intelligence, June 2023) - $4.4T annual gen-AI productivity uplift potential (McKinsey) - $200B+ annual AI infrastructure spending baseline 2025+ - 75% enterprise AI adoption target by 2026 (Gartner) - 32× AI economy growth: $40B (2022) → $1.3T (2032) The intersection — AI infrastructure applied to CRE Capital Markets workflow — is the opportunity Swarm & Bee LLC is built to capture. CRE shops cannot have a real AI strategy: capital structure can't justify the capex, talent funnel is structurally wrong (ML engineers don't work at brokerages), technical depth doesn't exist (no senior broker is debugging FSDP at 3am). The vendor model is the only realistic path for the existing industry. ## The vendor model (Phase 3) When M&M / JLL / Cushman / regional shops realize 12-18 months from now that they cannot build this themselves, vendor.defendable.eth becomes the only viable acquisition path. Two operational modes: - **Path A (current)**: Swarm & Bee LLC operates as a brokerage directly, expanding multi-state. - **Path B (Phase 3)**: Swarm & Bee LLC licenses the AI rails to existing licensed brokerages, who keep their license + clients while running on Atlas infrastructure. Service fee model. Both paths run in parallel as the franchise matures. ## Engagement For underwriting requests, deal handoffs, pass-or-proceed analysis, comp pulls, OM drafting, sensitivity studies, capital partner conversations, vendor licensing inquiries, recruiting (designated brokers + ML engineers), or general engagement: **Email**: build@swarmandbee.ai **Public site**: https://swarmandbee.ai **Architecture (open)**: https://github.com/SudoSuOps/atlas **Defendable standard**: https://defendable.eth.limo **Hedera anchor topic**: 0.0.10291838 (verifiable on hashscan.io) **ENS**: swarmandbee.eth · defendable.eth ## License Swarm & Bee LLC owns its compute, its Book of Business, its corpus, and its model weights. Architectural documentation (the Atlas repo) is publicly visible by intent — defendability is a public claim. Training corpus, model weights, and Book of Business data remain private. The Defendable standard (defendable.eth) is open-source by design — multiple implementations must agree on the same input. The standard is meant to be reproducible by anyone with the inputs. --- *This document is the canonical full-context manifest for Swarm & Bee LLC. Last updated 2026-05-04. For the concise version, see /llms.txt. For real-time deployment status, see https://swarmandbee.ai/status.*