Vercel platform products study (2026-10-17)

1. Why this doc

Young's direction 2026-10-16/17: Heart Fleet on all hosts; prod redesigned as the services lane (Docker + Medusa + POS + EVE + Hermes + Nuxt websites); pilot customers Young Le (youngle.au: Docs + Pulse) and YukonIT (yukonit.com.au: KB + EVE + Medusa + StoreFront); Mira + Tam to study Vercel templates for the product showcase. This doc registers (a) the Vercel template landscape against our product line, (b) MCP status for Hermes + opencode v2, (c) pilot mapping, (d) recommended product tiers. Companion: 02-architecture/prod-services-lane-redesign.md.

2. Vercel template landscape → our product line

We are already running the Vercel Docus template (docs-site/package.json: docus: latest, nuxt: ^4.4.8). The template library is not aspirational — it is the catalogue of what we can deploy today. Client demos = Vercel templates running our hosted services beneath them.

Verified live via GitHub API 2026-10-17 (stars, licence, push date):

Licence notes

• Docus MIT; nuxt-ui-pro family MIT; Medusa MIT; Agilo MIT; EVE Apache-2.0 — clean end-to-end, sellable.
• vercel/ai-chatbot = NOASSERTION — check NOTICE before packaging.
• vercel/platforms has no LICENSE file — pattern-reference only; do not copy code into client deliverables; write the pattern ourselves in the tenant template.
• Saleor Paper (FSL-1.1-ALv2) excluded — not resellable as a product; Medusa choice is what makes the bundle sellable.

Deliberately excluded

• The 14 EVE Slack/iMessage agent templates (they sell Slack agents, a different product).
• All Shopify/Salesforce/Wix storefronts (platform lock — the customer would own nothing).

3. Our product catalogue (5 products, all riding Nuxt + Docker)

1. KnowledgeBase — Nuxt/Docus site + vault-mirror pipeline + domain + email. Client writes markdown (or talks to an EVE agent); we host and publish. Showcase: docs.youngle.au itself.
2. BusinessOnline — Landing (Nuxt UI Pro) + domain + Office365 + contact/agentic chat button. Entry tier.
3. Commerce+POS — Medusa v2 storefront + Agilo POS on tablets + EVE receptionist, per-tenant stack on our Docker design. Showcase: the YukonIT pilot.
4. AIAgent hosting — EVE (or ai-chatbot) instance per client with tenant-scoped memory + HIL gates. Bundled with 2/3; standalone for shops who have a site already.
5. OpsDashboard — Nuxt UI Pro Dashboard for internal fleet dashboards (v1), then client ops dashboards (v2).

All five ride the same stack (Nuxt/Nuxt UI + our Docker/hosting design) — the showcase story is coherent: "everything you see on docs.youngle.au is the same machinery your business runs on."

4. MCP integration — Hermes + opencode v2

4.1 Status: NOT done. Gated pending Young's ruling.

• OPEN-DECISIONS.md item 4 — "The MCP ruling — keep or defer; the one conflict not resolvable by evidence" — Young owns it. Status: pending.
• Min's position (architectural): A2A carries work; MCP carries a single agent's access to its tools. They are not redundant and must not be merged. Phase 3 of DESIGN-OF-RECORD: (3.1) move the MCP tool list into the registry; (3.2) declare per-agent tool permission. MCP server reads the registry.
• Sol's position (empirical): production importers of arms/mcp_server.py = 0; measured usage = 0.03%. Building Phase 3 tooling around MCP repeats V2's "built-never-wired" defect. Defer Phase 3 to a gate — adopt A2A Part.data as the tool-call envelope, ship V3 with one real tool-consuming path, and let the need for MCP be demonstrated by a caller.
• STRUCTURE.md §8: core/arms/ and operation/mcp/ are deliberately gated empty directories — "until either a real consumer appears, or Young rules MCP stays." find . -name '*.py' returns 0 in .solos/; nothing built yet.

4.2 What is available today (V2 legacy, not wired)

• V2 arms/mcp_server.py exposes 10 tools inline (declared as code, not declaration) including verify_task — which bypasses the verification gate (the sharpest V2 defect: a caller can complete any working task with verified=True, no evidence inspected).
• arms/acp_surface.py also has 0 production importers.
• Hermes itself speaks A2A natively (a2a_call, a2a_discover, a2a_orchestrate, a2a_list, a2a_history) — the Brain's A2A client exists. MCP is a separate agent↔tool channel (stdio, spawned by the client).
• opencode v2.0.27 on all hosts (LAB, prod) — ARMS layer; A2A endpoints per agent per host.

4.3 How it would plug in (if ruled in)

• Tools live in cortex/ (one source); generated into the MCP tool list; per-agent tool permission declared. A refused tool call fails by declaration (not runtime).
• MCP server reads the registry; an agent loads only the skill/tool it needs for the task (Hermes pattern — V2 failed it: cards advertised 3–5 skills while every profile received all 67; empty skills: [] served as healthy).
• A2A and MCP are two interfaces, deliberately different — A2A carries tasks with a lifecycle (JSON-RPC over HTTP, one port per agent); MCP carries a single agent's access to its tools (stdio, spawned by the client). Merging them loses the task lifecycle.

4.4 Recommendation

Defer Phase 3 (MCP) to a gate. Do not build now. Adopt A2A Part.data as the tool-call envelope; ship V3 with one real tool-consuming path; let MCP be demonstrated by a caller before anything is built. This is the only path that escapes V2's "built-never-wired" lesson. Young's ruling required to change this.

5. Pilot customer mapping

5.1 Young Le — youngle.au

Services: Docs + Pulse websites.

• youngle.au — 000 (not resolving/serving) ⚠️ — apex landing page missing; this is the showcase gap.
• docs.youngle.au — 401 (auth-gated, by design) — Docus, live.
• pulse.youngle.au — 200, live.
• yukonit.youngle.au — 200 (vhost → :8012).
• Products delivered: P1 (Docs/KB) + P6 (Pulse/investing) already live. Gap = apex youngle.au landing (P5 marketing/landing) — the showcase Young is asking for.
• Pilot 1 = our own content pipeline as a service (near-zero marginal cost, high showcase value).

5.2 YukonIT — yukonit.com.au

Services: KB Website + EVE + Medusa + StoreFront.

• yukonit.com.au — 200, live, and not ours: <title>Domain Names & Web Hosting | Yukon IT Pty Ltd — an existing hosting business site.
• YukonIT is already a trading hosting business. We are not building a shop for a blank slate; we are adding KB + EVE + Medusa + storefront to an existing operator who will judge us by whether our stack beats what they already sell.
• Delivery set = P1 (KB) + P2 (storefront) + P4 (EVE) + hosting — exactly the bundle in business-plan-v2, and the pilot that proves the A$119 tier.
• Pilot 2 = the full bundle (A$119 tier target). YukonIT is the first tenant candidate (services-pricing-model-aud.md §7).

6. Recommended product tiers (based on Vercel examples + business plan)

6.1 Tier ladder (commerce bundle, AUD, from business-plan-v2)

6.2 Product tiers (showcase catalogue, from Vercel templates)

6.3 Add-on pricing (services-pricing-model-aud.md)

• Domain (.com.au): A$5/mo (A$2.08–3.33 wholesale).
• M365 email: A$15/user/mo (AU$10.50 Microsoft retail).
• Storage 20 GB: A$1.32/mo (Oracle Block Volume Balanced).
• Bandwidth: A$0.00 (10 TB/mo free).
• Compute (shared node): A$4–5/tenant/mo.
• AI: A$1.55–3.10/tenant/mo on shared GO (free-first ladder).

6.4 Margin

Solo A$119: variable cost A$19.45–23.25 → margin A$95.75–99.55 (80–84%). Break-even tenant #1, day one. Margin is labour (onboarding, support), not infra.

7. Honest gaps / not established

• No build exists for any tenant stack; all designs are specs.
• MCP ruling pending Young.
• youngle.au apex not resolving — DNS vs hosting unverified.
• YukonIT is an existing hosting operator; coexistence with their current platform is an integration question for Sol.
• Oracle PAYG excess (A$40–60/mo if credits expire) — needs Young's OCI console check.
• POS payments are manual path only; terminal integration Phase 2/3.
• EVE is beta (API churn) — pin versions.

8. Cross-links

• Prod redesign: 02-architecture/prod-services-lane-redesign.md.
• Business plan v2: 03-decisions/business-plan-v2-aud-allowance.md.
• Services pricing: 03-decisions/services-pricing-model-aud.md.
• Product catalogue: 03-decisions/product-catalogue-vercel-grounded.md.
• Vercel template offer catalogue: 01-research/products-from-vercel-templates.md.
• Docker design: 02-architecture/docker-systems-design.md.
• Commerce hosting: 02-architecture/COMMERCE-HOSTING-DESIGN-2026-10-13.md.
• Host topology: 02-architecture/host-topology-roles-2026-10-16.md.
• Open decisions: 03-decisions/OPEN-DECISIONS.md (MCP = item 4).

9. Register note

This doc consolidates the Vercel-template/product/MCP/pilot/tier thread into one research note so Min can register it on the vault. Companion architecture doc: 02-architecture/prod-services-lane-redesign.md.