Skip to content

Open-source platform for 3D AI agents. Turn text or a photo into a rigged, animated GLB avatar, give it an LLM brain, memory and a wallet, and embed it anywhere with one web component. Remote MCP server, x402 payments, Three.js, browser-native.

docs(growth): add event kits for IBM, OpenAI, OSF and NVIDIA, three directory submissions, and four startup program applications

nirholas93c177732c02
diff --git a/marketing/growth/applications/digitalocean-startups-memo.md b/marketing/growth/applications/digitalocean-startups-memo.md
new file mode 100644
index 000000000..9349f2681
--- /dev/null
+++ b/marketing/growth/applications/digitalocean-startups-memo.md
@@ -0,0 +1,62 @@
+# DigitalOcean Startups: go/no-go memo
+
+**Decision: NO-GO for now.** No application drafted.
+**Verified:** 2026-09-17 against https://www.digitalocean.com/startups and the codebase.
+**Revisit trigger:** three.ws ships a self-hostable product that belongs in the DigitalOcean
+Marketplace (see "What would make it a fit").
+
+---
+
+## What the program is (from the live page, 2026-09-17)
+
+| Item | Published terms |
+|---|---|
+| Benefit | 12 months of DigitalOcean credits, "amount varies by partner", generally a "$10,000 monthly limit"; separate GPU credit packages for select startups; Standard-tier paid support free for 15 months; DigitalOcean Marketplace access; partner discounts |
+| Eligibility | Raised $10 million or less; company website with a matching company email; a registered DigitalOcean team account on a business email; new to DigitalOcean credits; valid credit card on file; "AI-native startups" prioritised, service businesses ineligible |
+| Apply | https://sammydigitalocean.typeform.com/to/tZXAmt (choose "Other" without an accelerator or VC), or https://sammydigitalocean.typeform.com/hatch |
+
+## Does three.ws use DigitalOcean?
+
+**No.** Checked 2026-09-17:
+
+- `git grep -il -E "digitalocean|digital ocean|doctl|DO_API_TOKEN|ondigitalocean"` across the repo
+ (excluding `node_modules` and `package-lock.json`) matched only two marketing files that name the
+ program: [github-growth-surfaces.md](../github-growth-surfaces.md) and
+ [opportunities.csv](../opportunities.csv). No code, config, SDK, or deploy target.
+- `package-lock.json` contains no DigitalOcean package.
+- Production runs on Google Cloud: one Cloud Run service (`three-ws-api`), Cloud Run GPU model
+ services, and Cloud Scheduler, per [docs/ops/gcp-production.md](../../../docs/ops/gcp-production.md).
+
+## Why no-go
+
+1. **No honest usage story.** The program's value to DigitalOcean is a startup building on
+ DigitalOcean. Any application would have to describe a migration or a new workload that does not
+ exist. We never claim usage that does not exist.
+2. **It splits the production stack for no product gain.** Production, the GPU fleet, and the crons
+ are on Google Cloud, where the owner has already approved spending the Google for Startups credits
+ ([docs/ops/gcp-credits-plan.md](../../../docs/ops/gcp-credits-plan.md)). The workspace rules say to
+ prefer GCP and not to onboard new external paid providers without approval. A second cloud adds an
+ on-call surface, a second secrets store, and a second billing relationship.
+3. **The distribution upside needs a product we do not ship.** The strongest reason to join is the
+ DigitalOcean Marketplace. Marketplace listings are deployable apps (1-Click Droplets and similar).
+ three.ws is a hosted platform; its self-hostable pieces are not packaged for that.
+4. **Card on file and a new team account** are required up front, for a program whose credits we
+ would not use.
+
+## What would make it a fit
+
+Reopen this memo if any one of these becomes true:
+
+- A self-hostable three.ws component (for example a packaged 3D generation or rigging worker from
+ `workers/`) is published as a container that a DigitalOcean user could deploy, making a Marketplace
+ listing a real distribution channel.
+- The owner decides to run a real, non-duplicative workload on DigitalOcean GPUs, and approves the new
+ provider.
+
+## Owner-only inputs if reopened
+
+Funding raised (must be $10M or less), legal entity, business email owner, and the card for the team
+account.
+
+**Pipeline update:** [opportunities.csv](../opportunities.csv) row "DigitalOcean Startups distribution
+route" set to `no_go_no_usage`.
diff --git a/marketing/growth/applications/elevenlabs-startup-grants.md b/marketing/growth/applications/elevenlabs-startup-grants.md
new file mode 100644
index 000000000..9d2d42b0f
--- /dev/null
+++ b/marketing/growth/applications/elevenlabs-startup-grants.md
@@ -0,0 +1,132 @@
+# ElevenLabs Startup Grants: final application sheet
+
+**Status:** ready to submit **if** the owner confirms headcount under 25 and accepts the minors-policy
+disclosure below.
+**Verified:** 2026-09-17 against https://elevenlabs.io/startup-grants, the live `/voice` page,
+`/api/tts/catalog`, the Cloud Run service env, and the code. Background and the first draft are in
+the packet [elevenlabs-startup-grants.md](../../partner-packets/elevenlabs-startup-grants.md)
+(2026-09-16). This sheet supersedes the packet's answer text where the two differ.
+**Apply at:** https://elevenlabs.io/grants-application (terms checkbox, then the questions).
+**Deadline:** rolling; "applications are considered on a rolling basis", decisions "within one week".
+Pipeline target 2026-09-24.
+
+Nothing has been submitted.
+
+---
+
+## Owner-only inputs
+
+| Field | Note |
+|---|---|
+| Employee count | Must be under 25 "at the time of application". Not recorded anywhere in the repo |
+| Applicant name, role, business email | Use a three.ws address; personal email is disqualifying ("Only applications with valid business emails will be considered") |
+| Team description and founder bio | Not in the repo |
+| Legal entity and country | Terms exclude sanctioned jurisdictions |
+| Acceptance of the grant terms | Ticking the box accepts terms for the company, including ElevenLabs' licence to use the three.ws name and logo |
+| Prior ElevenLabs enterprise contract or prior grant | Neither is recorded in the repo; owner confirms |
+
+---
+
+## Eligibility check against the public criteria (2026-09-17)
+
+The grant provides 12 months of access, "33,000,000 characters" (about 680 hours of agent usage),
+stated as "$5,500+ in value", plus founder community access. Characters cannot be exchanged or resold.
+
+| ElevenLabs criterion (quoted) | Evidence | Verdict |
+|---|---|---|
+| "Monetized product use case" | The ElevenLabs lane is priced in code: `/api/tts/catalog` returns `billing: "credits"`, `usdPer1k: 0.3`; the `/voice` page reads "$0.50 per voice clone, $0.30 per 1,000 spoken characters". Platform-key clones charge the user's prepaid credit balance (`api/tts/eleven-clone.js`) | Meets |
+| "No short-term or one-off projects" | Generations recorded since 2026-06-11; public changelog | Meets |
+| Valid business email | three.ws domain | Meets (owner uses it) |
+| "startups or companies with less than 25 employees" | Not recorded | **Owner confirms** |
+| "No agencies or consulting firms" | Product company | Meets |
+| "Each company can only submit one application" | No prior application recorded | Meets as far as the repo shows |
+| Existing enterprise customers ineligible | No contract recorded | Meets as far as the repo shows |
+| "No projects for minors" | The platform is not built for children, but the live [Terms of Service](https://three.ws/legal/tos) say "You must be at least 13 years old to use the Service" | **Disclosure risk.** Do not describe the audience as adults-only. If asked, state the 13+ general policy honestly. The owner may choose to restrict the ElevenLabs voice features to 18+ before applying; that is a product decision, not a wording fix |
+
+**Go/no-go:** go, conditional on the headcount and the owner's call on the minors row.
+
+---
+
+## Live integration proof (verified 2026-09-17)
+
+| Claim | Evidence |
+|---|---|
+| `/voice` is live | `curl -o /dev/null -w "%{http_code}" https://three.ws/voice` returned `200` |
+| Shared ElevenLabs client exists | [`api/_lib/elevenlabs.js`](../../../api/_lib/elevenlabs.js), 219 lines: base URL `https://api.elevenlabs.io/v1`, default model `eleven_flash_v2_5`, `listVoices`, `createClonedVoice`, `deleteVoice`, per-request key resolution |
+| Models offered | `eleven_flash_v2_5`, `eleven_turbo_v2_5`, `eleven_multilingual_v2` (catalog and `TTS_MODELS`) |
+| Endpoints | `api/tts/eleven.js` (synthesis), `api/tts/eleven-clone.js` (instant cloning), `api/tts/eleven/voices.js`, `api/tts/eleven/library.js`, `api/agents/_id/voice.js` (per-agent voice) |
+| Lip-sync | Viseme analysis in `src/lip-sync-analyser.js`, imported by `src/agent-avatar.js`; the `<agent-3d>` element drives viseme morphs (`src/element.js`) |
+| **Lane state today** | `/api/tts/catalog`: ElevenLabs `available: false`, `byok: true`, `clone: true`, reason "Add your own ElevenLabs key below" |
+| **Platform key** | `node scripts/read-service-env.mjs '^ELEVENLABS' --names` returned "no variable on three-ws-api matches". There is no platform ElevenLabs key in production |
+
+The honest pitch is therefore: the integration is complete and metered, it runs today only on a user's
+own key, and the grant switches on the platform lane for every visitor.
+
+---
+
+## Application answers (paste as written)
+
+**Company one-liner**
+
+> three.ws turns a text prompt or photo into a rigged 3D AI agent that speaks, gestures, and reacts on
+> any web page.
+
+**What are you building?**
+
+> three.ws is an open-source platform for embodied AI agents. A user describes a character, gets a
+> textured and rigged 3D avatar, gives it a mind and a voice, and embeds it on any site with one web
+> component. ElevenLabs is integrated into that flow: speech synthesis, instant voice cloning, the
+> Voice Library, and per-agent voices, with Flash v2.5 as the default for low latency and Turbo v2.5
+> and Multilingual v2 selectable. The avatar's mouth follows the audio through viseme lip-sync, so the
+> voice becomes a face and body rather than an audio track beside a character.
+
+**How do you use (or plan to use) ElevenLabs?**
+
+> The integration is finished in production code and runs today on a bring-your-own-key basis: a user
+> adds their ElevenLabs key and synthesis and cloning bill their own account. We do not yet have a
+> platform ElevenLabs account, so visitors without a key hear our free voices instead. The grant would
+> switch on the platform lane so any visitor can give a 3D agent an ElevenLabs voice. We would use it
+> for default voices on public demo agents, voice cloning for creators, and a public browser demo in
+> which a cloned voice drives a rigged avatar in real time.
+
+**How do you make money / business model?**
+
+> ElevenLabs usage on the platform lane is already priced: $0.30 per 1,000 characters and $0.50 per
+> voice clone, charged to a prepaid credit balance, alongside paid high-quality 3D generation and
+> exports. Across the platform, 75,959 avatars and 3,831 agents exist and 632 widgets are embedded on
+> other sites. Every embedded agent that speaks with a premium voice is repeat, metered usage.
+
+**How do you expect to grow?**
+
+> Embeds and creators. The platform recorded 23,444 3D generations in the last 30 days. Each generated
+> character is a candidate for a voice, and each embed puts that voice on a site we do not own. Our
+> first milestone with the grant is making an ElevenLabs voice the default choice for newly created
+> public agents, measured by characters synthesized per week on the platform lane.
+
+**Team:** **OWNER INPUT** (names, roles, headcount).
+
+**Anything else**
+
+> We would welcome consideration for a future Startup Grants Demo Day with the step from a voice
+> agent to a visible digital human.
+
+Do not promise Demo Day 2026: https://demoday.elevenlabs.io/ lists the "Startup Grants Virtual Demo
+Day" on Wednesday, October 21st 2026, 6:00 PM (CEST), and a grant awarded now is unlikely to be in it.
+
+---
+
+## Figures used and how to refresh them
+
+| Figure | Value | Source | Captured |
+|---|---|---|---|
+| Avatars / agents / widgets | 75,959 / 3,831 / 632 | `curl -s https://three.ws/api/platform/stats` | 2026-09-17T02:58Z |
+| Generations, last 30 days | 23,444 | Production `forge_creations` (see the query in [google-cloud-ai-agents.md](./google-cloud-ai-agents.md)) | 2026-09-17 |
+| ElevenLabs pricing | $0.30 per 1,000 characters, $0.50 per clone | `/api/tts/catalog` and the `/voice` page | 2026-09-17 |
+| ElevenLabs characters synthesized | **Not measured.** BYOK calls bill the user's own ElevenLabs account and no platform key exists | n/a | n/a |
+
+**Relationship wording:** there is no relationship with ElevenLabs. Say "three.ws integrates the
+ElevenLabs API". After an award, "three.ws received an ElevenLabs Startup Grant", nothing stronger.
+
+**After an award:** add `ELEVENLABS_API_KEY` to the Cloud Run service with `--update-env-vars` (as a
+Secret Manager reference), confirm `/api/tts/catalog` flips the lane to `available: true`, then fix the
+Voice Lab copy that describes platform metering, and add a `data/changelog.json` entry.
diff --git a/marketing/growth/applications/google-cloud-ai-agents.md b/marketing/growth/applications/google-cloud-ai-agents.md
new file mode 100644
index 000000000..c160eb048
--- /dev/null
+++ b/marketing/growth/applications/google-cloud-ai-agents.md
@@ -0,0 +1,192 @@
+# Google Cloud: Partner Network, AI Agent Ecosystem Program, Marketplace
+
+**Status:** enrollment answers ready to paste. Marketplace A2A listing **not ready** (two blockers below).
+**Verified:** 2026-09-17 (live curl, Cloud Run logs, Google docs). Background, architecture, and the
+full business case live in the partner packet
+[google-cloud-ai-agents.md](../../partner-packets/google-cloud-ai-agents.md) (verified 2026-09-16);
+this file is the field-by-field paste sheet and only restates what changed or what a form needs.
+**Deadline:** none published. Pipeline target 2026-09-25.
+
+Nothing here has been submitted. Every submission is an owner action.
+
+---
+
+## Owner-only inputs
+
+| Field | Why only the owner can fill it |
+|---|---|
+| Legal entity name and state or country of incorporation | Not recorded in the repo. Google requires incorporation "in a supported region" |
+| Google account that becomes Partner Administrator | Accepts the Marketplace Vendor Agreement on the company's behalf |
+| Payments profile and bank details (Marketplace vendor account) | Financial identity |
+| Signatory for the Marketplace Vendor Agreement | Legal acceptance |
+| Headcount, revenue, funding stage (if the enrollment or program form asks) | Not recorded in the repo; never estimate |
+
+---
+
+## Blockers re-checked on 2026-09-17
+
+**1. Vertex AI is still billing-denied.** `gcloud logging read` on `three-ws-api` returned
+`vertex imagen failed, falling back: Lightning dunning decision is deny for project:
+projects/93741856042` at 2026-09-17T03:14:37Z, and two turnaround-view failures with the same
+reason seconds later. Google's listing requirements say A2A agents must "Use Google foundation or
+3rd party models hosted in Model Garden as a default configuration"
+([requirements](https://docs.cloud.google.com/marketplace/docs/partners/ai-agents)). The AI Agent
+Ecosystem Program also expects a Gemini or Model Garden model. Clear the billing hold first.
+
+**2. The live Agent Card is not a Marketplace-ready card.** `curl -s
+https://three.ws/.well-known/agent-card.json` on 2026-09-17:
+
+| Check | Live value | What the Marketplace copy needs |
+|---|---|---|
+| `url` | `https://three.ws/` (the website) | The A2A JSON-RPC endpoint. `POST https://three.ws/api/agents/a2a-paid` answers JSON-RPC (a bare `message/send` returned `-32602 Invalid params: message is required`), while `POST /api/a2a` returns 404 |
+| `version` | `1.5.1` | Match the platform (`/api/version` reports `1.5.2`, commit `477eb6f8c`, revision `three-ws-api-00438-cg2`) |
+| Auth declaration | Older `authentication: { schemes: ["bearer"] }` | A2A `securitySchemes` plus `security` |
+| `capabilities.extensions` | One per-call on-chain payment extension with `"required": true` | Remove it or mark it optional: a Marketplace customer is billed by Google, not per call on-chain |
+| `description` | Advertises crypto market data and names stablecoin chains | An enterprise description of the 3D skills only |
+| Storage | Served from the website | A JSON file in Cloud Storage, in the same project as the listing, attached with **Save and validate** in Producer Portal ([Agent Card doc](https://docs.cloud.google.com/marketplace/docs/partners/ai-agents/agent-card)) |
+
+Google's Agent Card doc recommends the card cover "skills, competence, models, and security". Build a
+separate Marketplace card rather than editing the public one, which other clients already read.
+
+---
+
+## Form 1: Google Cloud Partner Network enrollment
+
+**URL:** https://partners.cloud.google.com/enrollment (Google sign-in required, so the exact field
+labels could not be read logged out). Google's
+[Marketplace requirements](https://docs.cloud.google.com/marketplace/docs/partners/get-started)
+state what enrollment must establish: membership "in good standing", incorporation in a supported
+region, and a Partner Administrator. The answers below cover every item those docs name, plus the
+standard company-profile items.
+
+| Item | Answer |
+|---|---|
+| Company name | three.ws (legal entity: **OWNER INPUT**) |
+| Website | https://three.ws |
+| Business contact email | [email protected] (the partnership inbox named on `/partners`, per [docs/partners.md](../../../docs/partners.md)) |
+| Partner type, if asked | The software or ISV option. three.ws builds a product; it does not resell or deliver services |
+| Product description, if asked | three.ws turns a text prompt or photo into a textured, rigged, animation-ready 3D character and embeds it on any page with one web component. Production runs on Google Cloud Run, Cloud Scheduler, and a Cloud Run GPU fleet. |
+| Existing Google relationship, if asked | Member of Google Cloud for Web3 Startups (exact wording from [docs/partners.md](../../../docs/partners.md)). Do not write "Google Cloud Partner" anywhere until enrollment is accepted |
+| Google Cloud project for the listing | `aerial-vehicle-466722-p5` (production, region `us-central1`) |
+| Partner Administrator | **OWNER INPUT** |
+
+**After enrollment:** in Partner Network Hub, **View tasks → Partner tasks → Initiate onboarding
+your product to Marketplace**; accept the
+[Marketplace Vendor Agreement](https://cloud.google.com/terms/marketplace-vendor-agreement)
+(Partner Administrator); complete the **Project Info form** the Marketplace team sends, which grants
+Producer Portal access ([vendor steps](https://docs.cloud.google.com/marketplace/docs/partners/offer-products)).
+
+---
+
+## Form 2: Marketplace onboarding "supporting documentation"
+
+Google's vendor doc says this step may require "architecture diagrams or business inputs" and does not
+publish field names. Paste from these sources:
+
+| Input | Source |
+|---|---|
+| Architecture | Packet section "Architecture (for the diagram)": `three-ws-api` Cloud Run service, the Cloud Run GPU services, `three-ws-multiplayer`, Cloud Scheduler, Cloud Build, Vertex AI. Re-read the fleet with `gcloud run services list --region us-central1 --project aerial-vehicle-466722-p5` on the day of submission; the command timed out from this workspace on 2026-09-17, so the packet's 2026-09-16 read is the latest |
+| Cloud Scheduler job count | **120** jobs in us-central1 (`gcloud scheduler jobs list`, 2026-09-17), synced from **118** cron definitions in `vercel.json` |
+| Product and customer problem | Packet section "Marketplace business case" |
+| Hosting pattern | The "AI agents" pattern for the A2A listing type; Google lists it among its approved hosting patterns |
+| Production readiness | Google requires "production-ready (not alpha or beta)". Evidence table below |
+| Pricing model | **Free** first listing (Google supports Free, Subscription, Usage-based, Combined). Rationale in the packet |
+
+**Production evidence, refreshed 2026-09-17**
+
+| Metric | Value | Source | Captured |
+|---|---|---|---|
+| Generations recorded / completed (`status = 'done'`) | 40,506 / 34,738 since 2026-06-11 | Production `forge_creations` (read-only query) | 2026-09-17 |
+| Generations in the last 30 days | 23,444 (average 756 per day, peak day 1,960) | Same | 2026-09-17 |
+| Auto-rig jobs in the last 30 days | 3,929 | Same, `prompt = 'auto-rig'` | 2026-09-17 |
+| Avatars / agents / embedded widgets | 75,959 / 3,831 / 632 | `https://three.ws/api/platform/stats` | 2026-09-17T02:58Z |
+| Agent Card skills | 10 | `https://three.ws/.well-known/agent-card.json` | 2026-09-17 |
+| 90-day uptime | 99.85% overall | `https://three.ws/status` (packet) | 2026-09-16 |
+
+Re-run the counts on the day of submission:
+
+```bash
+node --env-file=.env.local --input-type=module -e "
+import { neon } from '@neondatabase/serverless';
+const sql = neon(process.env.DATABASE_URL);
+console.log(await sql\`select count(*)::int total,
+ count(*) filter (where status = 'done')::int done,
+ count(*) filter (where created_at > now() - interval '30 days')::int last30
+ from forge_creations\`);"
+curl -s https://three.ws/api/platform/stats
+```
+
+---
+
+## Form 3: Google Cloud AI Agent Ecosystem Program
+
+**Routes** (from Google's program announcement,
+[Build, deploy, and promote AI agents](https://cloud.google.com/blog/topics/partners/build-deploy-and-promote-ai-agents-through-the-google-cloud-ai-agent-ecosystem-program),
+published 2024-11-20): the application form at
+`https://docs.google.com/forms/d/10U-lTYO4J1fgSTjfKWl-EiqjFDnLLqaYvxtLVHSpUf0/viewform`, the partner
+representative, or the Marketplace sell page. The form returned **HTTP 401** to an unauthenticated
+request on 2026-09-17, and the Partner Network resource page
+(`https://partners.cloud.google.com/resources/google-cloud-ai-agent-program`) redirects to Google
+sign-in. Field labels are therefore unverified; the answers below cover the program's published
+qualification points.
+
+| Published criterion | three.ws answer | Verdict |
+|---|---|---|
+| Agent addresses a specific goal | Turns a text prompt or image into a textured, rigged 3D character or asset, and validates, inspects, and renders glTF/GLB models for other agents | Meets |
+| Uses a Gemini model or a third-party model from Model Garden | Gemini and Imagen lanes exist in code but are billing-denied today (see blocker 1) | **Fails until the billing hold is cleared** |
+| Deployed on Google Cloud | Cloud Run service `three-ws-api` in `us-central1`, plus Cloud Run GPU model services | Meets |
+| Uses or plans to migrate to Vertex AI services beyond the LLM | Imagen on Vertex for reference images; production 3D reconstruction and rigging already run on Cloud Run GPUs | Meets once Vertex billing is restored |
+| Program audience: ISV and services partners | ISV | Meets after Partner Network enrollment |
+
+**Paste-ready answers**
+
+**Company and agent, one sentence:**
+
+> three.ws is a 3D agent platform on Google Cloud whose A2A agent turns text or images into
+> textured, rigged, embeddable 3D characters, and validates and inspects glTF models for other agents.
+
+**What does the agent do and for whom?**
+
+> Enterprise product, marketing, and commerce teams want 3D assets and embodied agents but have no 3D
+> pipeline. Our agent takes a text prompt or reference image and returns a textured GLB, rigs humanoid
+> characters for animation, and returns an embeddable viewer. It also validates glTF/GLB files with
+> the Khronos validator and reports mesh, material, and texture statistics with optimization advice,
+> so other agents in a multi-agent workflow can check 3D output before using it.
+
+**How is it built on Google Cloud?**
+
+> Production is one Cloud Run service that serves the site, API, and agent endpoints, with Cloud
+> Scheduler running 120 jobs and a Cloud Run GPU fleet running the 3D reconstruction, rigging, and
+> motion models. Over the last 30 days the platform recorded 23,444 generations. Vertex AI provides
+> the Gemini and Imagen lanes.
+
+Send the last sentence only after blocker 1 is cleared, because today every Vertex call falls back.
+
+**What do you want from the program?**
+
+> Technical guidance to make a Model Garden model the agent's default, Gemini Enterprise validation
+> of the A2A agent, a Marketplace listing, and placement in the AI agent finder.
+
+**Relationship wording:** member of Google Cloud for Web3 Startups. Not a Google Cloud Partner until
+enrollment is accepted. No Marketplace listing exists.
+
+---
+
+## Form 4: Producer Portal (after Project Info form)
+
+Google's seven onboarding steps, each with the answer or the engineering prerequisite:
+
+| Step | Input | Ready? |
+|---|---|---|
+| 1. Create the AI agent and Agent Card | Marketplace copy of the card, fixes in the table above | **No: engineering** |
+| 2. Add the agent in Producer Portal | Listing type "AI Agents as a Service (A2A protocol)" | After step 1 |
+| 3. Add the Agent Card | Upload to Cloud Storage in `aerial-vehicle-466722-p5`, **Browse**, **Save and validate** | After step 1 |
+| 4. Product details | Name: three.ws 3D Agent. Short description: "Text or image to a textured, rigged, embeddable 3D model, plus glTF validation and inspection for agents." Link: https://three.ws/forge. Screenshot: [forge.png](../../partner-packets/images/forge.png) | Yes |
+| 5. Pricing (about 4 business days of review) | Free | Yes |
+| 6. Marketplace integration: account creation, Google account linking, sign-in | Not built. `git grep` over `api/`, `server/`, and `src/` on 2026-09-17 found no Cloud Commerce Procurement integration, and `api/auth/` has no Google sign-in provider | **No: engineering** |
+| 7. Publish after review | n/a | After 1 to 6 |
+
+## After submitting
+
+Record the Partner Network ID and any program receipt in
+[opportunities.csv](../opportunities.csv) row "Google Cloud AI Agents program and finder".
diff --git a/marketing/growth/applications/mongodb-for-startups-memo.md b/marketing/growth/applications/mongodb-for-startups-memo.md
new file mode 100644
index 000000000..500feb76c
--- /dev/null
+++ b/marketing/growth/applications/mongodb-for-startups-memo.md
@@ -0,0 +1,52 @@
+# MongoDB for Startups: go/no-go memo
+
+**Decision: NO-GO.** No application drafted.
+**Verified:** 2026-09-17 against https://www.mongodb.com/solutions/startups and the codebase.
+**Revisit trigger:** a real production workload moves to MongoDB Atlas for a product reason.
+
+---
+
+## What the program is (from the live page, 2026-09-17)
+
+| Item | Published terms |
+|---|---|
+| Benefit | Atlas credits and support in four tiers (Inspire, Grow, Innovate, Scale) by stage; credit amounts are not published on the page. Promo code has 12 months to activate and 12 months to use |
+| Eligibility (base tier) | Less than 7 years old and Series A or earlier; a single software product (no dev shops or agencies); not previously accepted; active company website and LinkedIn profile |
+| Product requirement | Must use **MongoDB Atlas** (new and existing Atlas users are eligible) |
+| Application fields | Company email, name, company name, company URL, LinkedIn profile, HQ country, funding stage |
+| Apply | The form on https://www.mongodb.com/solutions/startups, a VC or accelerator referral, or [email protected] |
+
+## Does three.ws use MongoDB?
+
+**No.** Checked 2026-09-17:
+
+- `git grep -il -E "mongodb|mongoose"` across the repo (excluding `node_modules`, `package-lock.json`,
+ `public/`, `dist/`) matched only: the two marketing files that name this program; the secrets
+ scanner `scripts/check-secrets.mjs` (a regex that detects `mongodb://` credentials in diffs); and
+ token-list JSON data files that list a tokenized asset named "MongoDB". None is a database
+ client.
+- `package-lock.json` has **0** `node_modules/mongodb` or `node_modules/mongoose` entries.
+- The actual data layer: Neon serverless Postgres through `@neondatabase/serverless`
+ ([`api/_lib/db.js`](../../../api/_lib/db.js)), Upstash Redis through `@upstash/redis` for rate limits
+ and caches, and an S3-compatible object store client through `@aws-sdk/client-s3`
+ ([`api/_lib/r2.js`](../../../api/_lib/r2.js)). The schema is SQL
+ ([`api/_lib/schema.sql`](../../../api/_lib/schema.sql)) with migrations in `api/_lib/migrations/`.
+
+## Why no-go
+
+1. **The program requires Atlas usage and we have none.** Applying would mean claiming, or promising,
+ a database we do not run. We never claim usage that does not exist.
+2. **Migrating to earn credits is backwards.** Production depends on relational features across
+ 388 SQL migration files (`ls api/_lib/migrations | wc -l`, 2026-09-17) and on Postgres-specific behaviour. Moving any of it to a document
+ database to qualify for credits would add risk with no user benefit.
+3. **The GTM upside depends on a genuine implementation story**, which is what the pipeline row itself
+ asked us to verify first ("Verify production usage ... before applying"). The verification failed.
+
+## What would make it a fit
+
+Only a product-driven reason, for example a new service whose natural store is a document database
+(vector search over large, schemaless agent memory) that the owner chooses to build on Atlas. At that
+point the application is a four-field form and this memo becomes the draft.
+
+**Pipeline update:** [opportunities.csv](../opportunities.csv) row "MongoDB for Startups GTM route" set
+to `no_go_no_usage`.
diff --git a/marketing/growth/kits/MKT-2026-09-IBM-EVENT.md b/marketing/growth/kits/MKT-2026-09-IBM-EVENT.md
new file mode 100644
index 000000000..f37c8eca9
--- /dev/null
+++ b/marketing/growth/kits/MKT-2026-09-IBM-EVENT.md
@@ -0,0 +1,213 @@
+# MKT-2026-09-IBM-EVENT: a user-group event where the 3D world is the venue and the demo
+
+**Kit status:** copy complete, verified 2026-09-17. **Blocked on a date** (campaigns.csv status
+`awaiting_date`) and therefore on the IBM Community event page, which is the registration link every
+post points at.
+
+## At a glance
+
+| Field | Value |
+|---|---|
+| Campaign id | `MKT-2026-09-IBM-EVENT` |
+| Planned publish date | 2026-09-22, as the T-14 announcement; the event itself runs 14 days later |
+| Summary | The second Three.ws User Group session meets inside three.ws: attendees arrive as avatars, talk with spatial voice, and watch an enterprise agent get a rigged 3D body live |
+| Audience | enterprise builders and community |
+| Primary channel | IBM Community (the user group's event page) |
+| Secondary channels | X, Telegram community, three.ws news |
+| Proof | https://three.ws/play |
+| Tracked link, IBM event page | `https://three.ws/play?utm_source=ibm-community&utm_medium=partner&utm_campaign=mkt-2026-09-ibm-event&utm_content=event-page` |
+| Tracked link, X | `https://three.ws/play?utm_source=x&utm_medium=social&utm_campaign=mkt-2026-09-ibm-event&utm_content=t-1` |
+| Tracked link, Telegram | `https://three.ws/play?utm_source=telegram&utm_medium=community&utm_campaign=mkt-2026-09-ibm-event&utm_content=announce` |
+| The single CTA | Register and bring an avatar or build to show |
+| Partner ask | Provide one product speaker and amplify the IBM event page |
+| KPI | registrations and live attendees |
+
+`{{IBM_EVENT_URL}}` below is the event page URL IBM Community generates when the event is created in
+the group. The X anchor links it (registration is the CTA); an IBM-domain URL carries no UTM.
+
+## Audit: what already exists
+
+| Channel | Existing asset | State | How this kit uses it |
+|---|---|---|---|
+| Event format and costing | [docs/ibm-next-event.md](../../../docs/ibm-next-event.md) | Proposal (recommends a week-long Forge-Off Open) | This campaign is the one-hour in-world session from the [opportunity register](../opportunities.md) ("Give an enterprise agent a 3D body, live"). If the owner picks the Forge-Off Open instead, keep this kit's channels and swap the run of show |
+| Partner relationship note (speaker, calendar listing, June follow-ups) | [templates.md, IBM relationship note](../templates.md) | Ready | Sent first, unchanged |
+| First meetup X copy | [docs/x-posts/event-x-posts.md](../../../docs/x-posts/event-x-posts.md), [x-meetup-posts.md](../../../docs/x-posts/x-meetup-posts.md) | Posted in August | Pattern only (IBM first, absolute clock time, never "in N hours"). Its Granite and payment claims are **not** reused: production has no `WATSONX_*` credentials |
+| First meetup blog (IBM Community) | [docs/ibm-community-blog-meetup-jessica.md](../../../docs/ibm-community-blog-meetup-jessica.md) | Published | Linked from the event page as the precedent |
+| Event page copy, X anchor for event two, Telegram, news, amplification ask, clip script, T-1, run of show, recap | none | Missing | Written below |
+
+## IBM Community event page (primary channel)
+
+**Title:** Three.ws User Group, session two: give an enterprise agent a 3D body, live
+
+**Description:**
+
+```text
+Most user groups meet on a video call. This one meets inside the product.
+
+Open three.ws/play in a browser and you arrive as a 3D avatar in a shared world. Walk up to someone and you hear them through spatial voice. Nothing to install, no account needed to look around.
+
+In one hour:
+1. Live build: a one-sentence prompt becomes a rigged 3D avatar, and one <agent-3d> tag puts it on a blank web page.
+2. Guest segment: {{SPEAKER_NAME}}, {{SPEAKER_ROLE}}, on {{SPEAKER_TOPIC}}.
+3. Community showcase: bring an avatar or a build and show it to the room.
+4. Open Q&A. The skeptical questions are the most useful ones.
+
+When: {{EVENT_TIME_UTC}}
+Where: https://three.ws/play?utm_source=ibm-community&utm_medium=partner&utm_campaign=mkt-2026-09-ibm-event&utm_content=event-page
+Before you come: make your avatar at https://three.ws/create so you arrive as yourself.
+
+How the first in-world meetup went: see the group blog post "The First Three.ws User Group Meetup: How It Actually Went".
+
+three.ws is an IBM Business Partner. This is a community event hosted by three.ws in the Three.ws User Group; it is not an IBM product event.
+```
+
+If no IBM speaker is confirmed by T-7, delete item 2 and renumber; never list an unconfirmed name.
+
+## X
+
+**Media:** [docs/media/meetup-banner-x.png](../../../docs/media/meetup-banner-x.png) until a fresh
+in-world capture exists (headless capture of `/play` crashed on 2026-09-17; record the world in a real
+browser for the T-7 clip and pull a still from it). No IBM mark on any image.
+**Alt text:** Avatars gathered in the three.ws plaza for a Three.ws User Group meetup.
+
+**Anchor, T-14** (171 weighted characters, measured with a placeholder IBM URL):
+
+```text
+The Three.ws User Group on IBM Community meets inside a live 3D world. Make an avatar from one sentence, walk in, and show the room what you built: {{IBM_EVENT_URL}}
+```
+
+**Thread (optional):**
+
+2/
+```text
+The hour: a prompt becomes a rigged avatar and goes onto a blank page with one tag, then a community showcase, then open Q&A. Spatial voice, so you hear people from where they stand.
+```
+
+3/
+```text
+The first in-world meetup ran in August. The world is open to anyone with the link, so bring a friend who has never heard of a user group. Time: {{EVENT_TIME_UTC}}
+```
+
+No `@IBM` tag: the voice rule is to tag IBM when the thing runs on watsonx, and Granite is not running
+in production today.
+
+## Telegram community
+
+```text
+The Three.ws User Group on IBM Community meets again, and again it meets inside three.ws.
+
+{{EVENT_TIME_UTC}}. Open the world in a browser, arrive as your avatar, and talk with spatial voice. We build an agent's 3D body live, then hand the floor to the community: bring an avatar or a build to show.
+
+Register on IBM Community: {{IBM_EVENT_URL}}
+Make your avatar first: https://three.ws/create?utm_source=telegram&utm_medium=community&utm_campaign=mkt-2026-09-ibm-event&utm_content=announce
+```
+
+## three.ws news blurb
+
+The Three.ws User Group on IBM Community holds its second session on {{EVENT_TIME_UTC}}, inside the
+three.ws world rather than on a call. Attendees arrive as avatars, watch an agent get a rigged 3D body
+built live, and show their own builds to the room. Registration is on the group's IBM Community event
+page.
+
+## Partner messages
+
+**1. Speaker and calendar listing:** send the existing
+[IBM relationship note](../templates.md) unchanged, in the existing thread with the IBM marketing
+contact from the 2026-06-15 and 2026-06-18 meetings (intake named in
+[docs/ibm-visibility-map.md](../../../docs/ibm-visibility-map.md), Tier 1 rows 3 and 4). Send the Agent
+Connect `APP_ID` request separately, as the template says.
+
+**2. Amplification, the day the event page is live (same thread):**
+
+```text
+Subject: Three.ws User Group session two is live on IBM Community
+
+The event page is up: {{IBM_EVENT_URL}}
+
+It runs {{EVENT_TIME_UTC}} inside the three.ws world, the same format as the August meetup: attendees arrive as avatars, a live build, a community showcase, and Q&A.
+
+One ask: could the IBM Community team feature the event on the community calendar or newsletter, and could an IBM social account repost our announcement ({{X_POST_URL}})? Attached: a 16:9 image, a 20-second clip of the world, and alt text for both. We will send the recap with attendance and replay within 24 hours of the session.
+```
+
+Contact names are not in the repo; address it to the existing thread, do not guess a name.
+
+## Event package
+
+`{{EVENT_TIME_UTC}}` is the confirmed start written as `YYYY-MM-DD HH:MM UTC`; add one US Pacific and
+one US Eastern time beside it on the event page. It is filled when the owner confirms the date, then
+the in-world schedule is set with `npm run event:schedule -- --start <ISO> --duration 60 --apply`
+(writes `public/event.json`; it needs a deploy to go live, which is owner-gated).
+
+**T-7 demo clip, 20 seconds:**
+
+| Time | Shot | Route |
+|---|---|---|
+| 0:00 to 0:05 | Type one sentence; the avatar generates | https://three.ws/create |
+| 0:05 to 0:10 | The rigged avatar idles and walks | https://three.ws/create result |
+| 0:10 to 0:16 | Walk into the world as that avatar; other avatars in frame | https://three.ws/play |
+| 0:16 to 0:20 | End card: "Three.ws User Group, {{EVENT_TIME_UTC}}, register on IBM Community" | card |
+
+**T-1 reminder** (197 weighted characters with the token in place):
+
+```text
+Tomorrow at {{EVENT_TIME_UTC}}: the Three.ws User Group meets inside three.ws. Open the world in a browser, arrive as your avatar, talk with spatial voice. Free, no install: https://three.ws/play?utm_source=x&utm_medium=social&utm_campaign=mkt-2026-09-ibm-event&utm_content=t-1
+```
+
+**Live run of show (60 minutes):**
+
+| Minute | Segment | Owner |
+|---|---|---|
+| T-30 | Doors open in the world; host in the plaza; moderator checks spatial voice | Host, moderator |
+| 0 to 5 | Welcome, house rules, where Q&A goes | Host |
+| 5 to 20 | Live build: prompt to rigged avatar at `/create`, then one `<agent-3d>` tag on a blank page | Host |
+| 20 to 35 | Guest segment (only if confirmed) | IBM speaker |
+| 35 to 50 | Community showcase: attendees walk up and present | Moderator calls names |
+| 50 to 60 | Q&A, next date, where the recap will be posted | Host |
+
+Fallback: a pre-generated avatar staged in case a generation worker is slow; the build segment never
+waits on a live queue.
+
+**T+1 recap skeleton:**
+
+```text
+Three.ws User Group, session two: the recap.
+
+Registered: {{REGISTRATIONS}}
+Peak avatars in the world: {{PEAK_CONCURRENCY}}
+Built live: {{BUILD_RESULT_URL}}
+Showcased by the community: {{SHOWCASE_LINKS}}
+Replay: {{REPLAY_URL}}
+Next session: {{NEXT_EVENT_URL}}
+```
+
+| Field | Source |
+|---|---|
+| `{{REGISTRATIONS}}` | IBM Community event page registrant count (group admin view) |
+| `{{PEAK_CONCURRENCY}}` | multiplayer server presence during the window (Cloud Run logs for `three-ws-multiplayer`); the first-meetup recap used the same measure |
+| `{{BUILD_RESULT_URL}}` | the avatar page generated in the build segment |
+| `{{SHOWCASE_LINKS}}` | links the presenters shared in chat |
+| `{{REPLAY_URL}}` | the uploaded recording |
+| `{{NEXT_EVENT_URL}}` | next event page, or omit the line |
+| Landing sessions (scorecard only) | web analytics, `utm_campaign=mkt-2026-09-ibm-event` |
+
+## Claims check
+
+| Claim | Proof | Verified |
+|---|---|---|
+| Three.ws User Group exists on IBM Community and we moderate it | group page returns 200; [docs/partners/opportunities.md](../../../docs/partners/opportunities.md) "Live and working" | 2026-09-17 |
+| A first in-world meetup already ran | [docs/ibm-community.md](../../../docs/ibm-community.md), blog posts 3 and 4 | repo |
+| `/play` is a live shared 3D world | https://three.ws/play returns 200 | 2026-09-17 |
+| Spatial voice | `multiplayer/src/rooms/WalkRoom.js` voice signaling (`voice-state`, `voice-signal`) | 2026-09-17 |
+| One sentence to rigged avatar at `/create` | [docs/ibm-next-event.md](../../../docs/ibm-next-event.md) capability table; `/create` returns 200 | 2026-09-17 |
+| No account needed to look around | `/play` welcome dialog text ("look around with no wallet") in `marketing/the-first-19-weeks/images/play.png` | repo |
+| three.ws is an IBM Business Partner | [docs/partners.md](../../../docs/partners.md) | repo |
+| No event currently scheduled in-world | https://three.ws/event.json returns the no-event state | 2026-09-17 |
+
+Dropped: "agents think on IBM Granite" (`/api/ibm/galaxy` returns `watsonx_not_configured` on
+2026-09-17), attendance numbers from the first meetup (not needed for the ask), and the commemorative
+wearable (only true if a souvenir id is set in `public/event.json` for this event; add it back only then).
+
+## Owner does
+
+Pick the date, send the IBM relationship note in the existing thread, and create the event in the
+Three.ws User Group with the description above.
diff --git a/marketing/growth/kits/MKT-2026-09-OPENAI-STUDIO.md b/marketing/growth/kits/MKT-2026-09-OPENAI-STUDIO.md
new file mode 100644
index 000000000..b9e081cf1
--- /dev/null
+++ b/marketing/growth/kits/MKT-2026-09-OPENAI-STUDIO.md
@@ -0,0 +1,143 @@
+# MKT-2026-09-OPENAI-STUDIO: create a real 3D object inside an AI conversation
+
+**Kit status:** copy complete, verified 2026-09-17 with a live keyless generation. Not blocked on
+engineering. campaigns.csv status `awaiting_submission`: the Showcase packet and the Developer
+Community post are written and unsent.
+
+## At a glance
+
+| Field | Value |
+|---|---|
+| Campaign id | `MKT-2026-09-OPENAI-STUDIO` |
+| Publish date | 2026-09-29 |
+| Summary | One sentence in ChatGPT returns a textured, rotatable 3D model with an AR link, from a free keyless MCP connector |
+| Audience | AI builders |
+| Primary channel | OpenAI Developer Community |
+| Secondary channels | X, LinkedIn, Telegram community, three.ws news |
+| Proof | https://three.ws/openai |
+| Tracked links | X `https://three.ws/openai?utm_source=x&utm_medium=social&utm_campaign=mkt-2026-09-openai-studio&utm_content=anchor`; LinkedIn `...utm_source=linkedin&utm_medium=social&...&utm_content=post`; Telegram `...utm_source=telegram&utm_medium=community&...&utm_content=announce`; forum `...utm_source=openai-community&utm_medium=community&...&utm_content=forum-post` |
+| The single CTA | Try the keyless 3D Studio and share the result |
+| Partner ask | Showcase or directory placement and partner-GTM introduction |
+| KPI | tool calls and completed generations |
+
+## Audit: what already exists
+
+| Channel | Existing asset | State | How this kit uses it |
+|---|---|---|---|
+| OpenAI Developer Community (primary) | [docs/openai-community-3d-studio-post.md](../../../docs/openai-community-3d-studio-post.md) | Draft, unposted | **Is** the primary post. Before posting, replace its tool list with the live eleven (see claims check) and add the forum UTM to its first three.ws link |
+| X and LinkedIn, partner-status beat | [marketing/openai-select-partner/social-copy.md](../../openai-select-partner/social-copy.md) | Posted beat (Select Partner announcement) | Not reposted. Its wording rules (full "OpenAI Select Partner", no endorsement) govern this kit |
+| X, product demo quote post | same file, @nichxbt quote tweet with video | Written | Optional companion from the personal account on the same day |
+| Showcase Gallery submission | [prompts/store-submissions/_generated/openai-showcase-submission.md](../../../prompts/store-submissions/_generated/openai-showcase-submission.md) | Ready to paste | Partner ask, part 1 |
+| Listing channel map | [docs/openai-listing-channels.md](../../../docs/openai-listing-channels.md) | Current | Source for intake paths |
+| Media | [marketing/openai-select-partner/cards](../../openai-select-partner/cards/), `https://three.ws/partners/openai/social-card-studio.png` (200), [partner-packets/images/openai-page.png](../../partner-packets/images/openai-page.png) | Exist | Anchor image: the Studio card |
+| X anchor on the Studio mechanism, LinkedIn Studio post, Telegram, news, partner-GTM note | none | Missing | Written below |
+
+## X
+
+**Media:** `https://three.ws/partners/openai/social-card-studio.png` (3200x1800).
+**Alt text:** three.ws 3D Studio card: a textured 3D model rendered inline in a ChatGPT conversation,
+with the note that the connector is free and keyless.
+
+**Anchor** (174 weighted characters):
+
+```text
+Ask ChatGPT for a ceramic robot and a textured 3D model comes back inline: rotate it, rig it, place it in your room. Eleven keyless tools, no account: https://three.ws/openai?utm_source=x&utm_medium=social&utm_campaign=mkt-2026-09-openai-studio&utm_content=anchor
+```
+
+**Thread (optional):**
+
+2/ (154)
+```text
+The connector is one MCP server at three.ws/api/mcp-studio. No auth header, no key, no wallet. Any MCP client can call the same eleven tools ChatGPT does.
+```
+
+3/ (188)
+```text
+Every result also returns structured content: a GLB URL, a viewer URL, an AR URL, and the concept image it was sculpted from, so a client can render it natively instead of printing a link.
+```
+
+4/ (175)
+```text
+look_at_model renders a GLB from several angles and hands the frames back to the model, so the conversation can judge its own output and call refine_model on what looks wrong.
+```
+
+No hashtags on this anchor: it is a product post, not the partner announcement, so the voice contract
+applies. No OpenAI account is tagged; no OpenAI X handle is recorded in the repo.
+
+## LinkedIn
+
+```text
+A chat assistant can describe a product for as long as you like. Until recently it could not hand you one.
+
+three.ws 3D Studio is a free MCP connector for ChatGPT. Type "a small ceramic robot figurine" and the conversation returns a textured 3D model you can rotate inline, a link that places it in your room in AR, and a downloadable GLB that opens in any 3D tool. The same connector can add a humanoid skeleton, refine a result, and render a model from several angles so the assistant can judge its own output before you do.
+
+It needs no account, no API key, and no payment. It is also a plain MCP server, so any MCP client can call the same eleven tools, and the implementation is open source under Apache-2.0.
+
+three.ws is an OpenAI Select Partner in the OpenAI Partner Network. It is not an OpenAI product and is not endorsed by OpenAI beyond that designation.
+
+Try it and share what you make: https://three.ws/openai?utm_source=linkedin&utm_medium=social&utm_campaign=mkt-2026-09-openai-studio&utm_content=post
+```
+
+(151 words.)
+
+## Telegram community
+
+```text
+3D inside ChatGPT, free and keyless.
+
+Connect the three.ws 3D Studio connector and ask for any object. You get a textured model you can rotate right in the chat, an AR link to place it in your room, and the GLB to download. It can rig characters and refine results too.
+
+No account, no key, no payment. Any MCP client works, not only ChatGPT.
+
+Post what you make in the community chat: https://three.ws/openai?utm_source=telegram&utm_medium=community&utm_campaign=mkt-2026-09-openai-studio&utm_content=announce
+```
+
+## three.ws news blurb
+
+The three.ws 3D Studio connector puts real 3D creation inside a ChatGPT conversation: one sentence
+returns a textured model to rotate inline, an AR link, and a downloadable GLB. It exposes eleven
+keyless tools over MCP, so any MCP client can use it with no account or key. We also published a
+technical write-up of the integration on the OpenAI Developer Community: {{FORUM_POST_URL}}.
+
+## Partner messages
+
+**1. Showcase Gallery:** submit the ready
+[Showcase packet](../../../prompts/store-submissions/_generated/openai-showcase-submission.md) at
+`https://openai.com/form/showcase-submission` (the form the developers.openai.com showcase page links;
+it blocks non-browser clients, so it was not fetched here). The final two attestations need a person
+authorized to accept the Showcase Gallery Program Agreement.
+
+**2. Partner-GTM introduction**, sent through the OpenAI Partner Network partner portal (the
+authenticated partner record; no named OpenAI contact exists in the repo):
+
+```text
+Subject: three.ws (OpenAI Select Partner): a measured 3D Studio story for partner GTM
+
+three.ws 3D Studio is a free, keyless MCP connector that lets ChatGPT create a textured, rotatable 3D model inline, with an AR placement link and a downloadable GLB. We published the build on the OpenAI Developer Community ({{FORUM_POST_URL}}) and submitted the open-source project to the Showcase Gallery on {{SHOWCASE_SUBMITTED_DATE}}.
+
+Seven-day numbers from the launch: {{TOOL_CALLS_7D}} tool calls and {{GENERATIONS_7D}} completed generations.
+
+Two asks: could you introduce us to the partner GTM contact for Select Partners building on the Apps SDK, and confirm that three.ws appears in the Partner Locator? If it is absent, we would like to know what the profile is missing.
+```
+
+Send it only after the seven-day numbers exist; do not send it with the fields empty.
+
+## Claims check
+
+| Claim | Proof | Verified |
+|---|---|---|
+| Eleven tools, keyless, no auth | `tools/list` on `POST https://three.ws/api/mcp-studio` returned 11: `forge_free`, `text_to_avatar`, `mesh_forge`, `rig_mesh`, `forge_avatar`, `refine_model`, `check_job`, `create_agent_persona`, `get_agent_persona`, `persona_say`, `look_at_model` | 2026-09-17 |
+| A prompt returns a GLB, viewer, AR link, and concept image | live `tools/call forge_free` with "a small ceramic robot figurine": `structuredContent` carried `glbUrl`, `viewerUrl`, `arUrl`, `referenceImageUrl` | 2026-09-17 |
+| Renders inline in ChatGPT | Apps SDK widget described in [docs/openai-community-3d-studio-post.md](../../../docs/openai-community-3d-studio-post.md) and [docs/mcp-studio.md](../../../docs/mcp-studio.md) | repo |
+| Can rig and refine | `rig_mesh`, `text_to_avatar`, `refine_model` in the live tool list | 2026-09-17 |
+| `look_at_model` renders several angles for the model | live tool list; `POST /api/3d/look` returned per-angle frame URLs | 2026-09-17 |
+| OpenAI Select Partner, not endorsed | [docs/partners.md](../../../docs/partners.md); `/openai` meta description | 2026-09-17 |
+| Apache-2.0 | `gh repo view nirholas/three.ws` license | 2026-09-17 |
+
+Dropped: generation time ("about a minute") because it varies by lane and queue. Drift found: the
+community draft names ten tools (no `look_at_model`) while saying eleven; fix before posting.
+
+## Owner does
+
+Post the OpenAI Developer Community draft (with the eleven-tool fix), then publish the X anchor with
+the Studio card and submit the Showcase form.
diff --git a/marketing/growth/kits/MKT-2026-09-OSF.md b/marketing/growth/kits/MKT-2026-09-OSF.md
new file mode 100644
index 000000000..3274c7c18
--- /dev/null
+++ b/marketing/growth/kits/MKT-2026-09-OSF.md
@@ -0,0 +1,213 @@
+# MKT-2026-09-OSF: Open Source Friday, the community shapes the live contribution
+
+**Kit status:** copy complete, verified 2026-09-17. **Blocked on two owner steps** before anything
+posts: the stream date is not booked, and there is no open `good first issue` for anyone to claim.
+
+## At a glance
+
+| Field | Value |
+|---|---|
+| Campaign id | `MKT-2026-09-OSF` |
+| Planned publish date | 2026-09-17 (campaigns.csv); slips to the day the booking is confirmed |
+| Summary | GitHub approved three.ws for Open Source Friday; the community picks the rig convention we implement, test, and open as a PR on stream |
+| Audience | open-source builders |
+| Primary channel | X |
+| Secondary channels | GitHub Discussions, Telegram community |
+| Proof | https://github.com/githubevents/open-source-friday/issues/254 (open, `approved` label) |
+| Tracked landing link | `https://three.ws/rig-doctor?utm_source=x&utm_medium=social&utm_campaign=mkt-2026-09-osf&utm_content=anchor` |
+| The single CTA | Claim or propose a good-first issue before the stream |
+| Partner ask | Confirm the date and include the visual demo in official promotion |
+| KPI | claimed issues and confirmed stream date |
+
+The proof is a GitHub-owned URL, so it carries no UTM. Every link we control does.
+
+## Audit: what already exists
+
+| Channel | Existing asset | State | How this kit uses it |
+|---|---|---|---|
+| Stream plan, runsheet, promotion sequence | [docs/open-source-friday-plan.md](../../../docs/open-source-friday-plan.md) | Complete; post plan described but T-7, T-1, T-0, T+1 copy never written | Runsheet reused below; missing copy written here |
+| X, approval beat | [docs/announcements/open-source-friday.post.txt](../../../docs/announcements/open-source-friday.post.txt) and its pack | Drafted, never posted | Same news beat as this anchor. **Post one, not both.** This kit's anchor carries the campaign CTA, so prefer it; the pack remains the Rig Doctor feature post |
+| X, reply to the host's thread | [docs/x-posts/osf-issue-reply.md](../../../docs/x-posts/osf-issue-reply.md) | Written for 2026-08-25 | Stale (application beat). Not reused |
+| Partner message, booking reply on #254 | [templates.md, "GitHub Open Source Friday booking follow-through"](../templates.md#github-open-source-friday-booking-follow-through) | Ready, only after booking | Reused as message 1 |
+| Media | `public/announce/img/rig-doctor-hero.webp` (captured by `npm run announce:media`) | Exists | Anchor image |
+| GitHub Discussions, Telegram, T-7 clip script, T-1, recap, promotion-assets message | none | Missing | Written below |
+
+## Blockers, in order
+
+1. **Book the slot.** Issue #254 still has only GitHub's 2026-09-01 "Please select a date" comment
+ (checked with `gh issue view 254` on 2026-09-17). Booking page: `https://gh.io/osf-booking`
+ (redirects to Microsoft Bookings; curl cannot read it, a browser can).
+2. **Open fresh starter issues.** The six curated issues (#110 to #115) are all **closed**, and the
+ repo had **zero open issues** on 2026-09-17. The anchor therefore asks people to *propose* a
+ convention in Discussions, and the owner (or engineering) should open two or three new
+ `good first issue` rig tickets from those proposals before T-7, so the T-7 post has something to
+ claim. Until then, do not write "claim an open issue" anywhere.
+
+## X
+
+**Media:** `public/announce/img/rig-doctor-hero.webp`.
+**Alt text:** three.ws Rig Doctor: drop in any humanoid GLB and it names the skeleton convention, then
+lists joint by joint which bones will animate and which will not move.
+
+**Anchor** (177 weighted characters, `post-tweet.mjs --dry-run`):
+
+```text
+Bring the skeleton our retargeter cannot read. On GitHub's Open Source Friday we map one new rig convention live, write its test, and open the PR on air: https://three.ws/rig-doctor?utm_source=x&utm_medium=social&utm_campaign=mkt-2026-09-osf&utm_content=anchor
+```
+
+**Thread (optional, replies to the anchor):**
+
+2/ (197)
+```text
+Rig Doctor already fingerprints 15 skeleton conventions, from Mixamo and VRM to MikuMikuDance, Kinect and 3ds Max Biped, and lists joint by joint what will not move. The file never leaves your tab.
+```
+
+3/ (188)
+```text
+All six starter issues we opened for this are closed. So the next one comes from you: name the rig, link a sample GLB, and we pick the convention we implement live. https://github.com/nirholas/three.ws/discussions/categories/ideas?utm_source=x&utm_medium=social&utm_campaign=mkt-2026-09-osf&utm_content=thread-ideas
+```
+
+4/ (168)
+```text
+A new convention is one mapping entry in src/glb-canonicalize.js plus a test case, and the guide goes from clone to open PR in about 15 minutes: https://github.com/nirholas/three.ws/blob/main/docs/first-contribution.md
+```
+
+No tags: the approval pack's rule stands (do not tag @github or the hosts before the date is booked).
+After booking, the T-0 post is a quote of GitHub's own announcement, per plan section 5.
+
+## GitHub Discussions (category: Ideas)
+
+**Title:** Pick the rig convention we implement live on GitHub's Open Source Friday
+
+```text
+three.ws is approved for GitHub's Open Source Friday (githubevents/open-source-friday#254). The live segment is a real contribution: teach the retargeter a bone-naming convention it does not know, write the test, open the PR on stream.
+
+Rig Doctor (https://three.ws/rig-doctor) already recognises 15 conventions. Tell us the 16th.
+
+Reply with:
+1. The tool or engine the rig comes from.
+2. Three or four real bone names (hips, left upper arm, spine).
+3. A link to a sample GLB you are allowed to share, or a Rig Doctor screenshot of what fails to map.
+
+The best-documented proposals become good-first issues before the stream, and anyone can claim one and pair with us on air. How a mapping lands: docs/first-contribution.md.
+```
+
+## Telegram community
+
+```text
+three.ws is approved for GitHub's Open Source Friday livestream.
+
+The centrepiece is a real contribution done on air: we teach the avatar retargeter a skeleton convention it cannot read yet, write the test, and open the PR live. You choose the convention. Drop a rig in Rig Doctor, and if joints fail to map, post the rig in GitHub Discussions (Ideas).
+
+Rig Doctor: https://three.ws/rig-doctor?utm_source=telegram&utm_medium=community&utm_campaign=mkt-2026-09-osf&utm_content=announce
+Stream date follows as soon as it is booked.
+```
+
+## three.ws news blurb
+
+GitHub approved three.ws for Open Source Friday, its weekly livestream with open-source maintainers.
+The stream's centrepiece is a contribution made on air: the community proposes a rig convention that
+Rig Doctor cannot read yet, and we implement it, test it, and open the pull request live.
+
+## Partner messages
+
+**1. After booking, on issue #254:** use the existing
+[booking follow-through reply](../templates.md#github-open-source-friday-booking-follow-through)
+verbatim, with the booked date.
+
+**2. The promotion-assets ask, on issue #254, the Monday before the stream (T-4):**
+
+```text
+Assets for the official promotion, ready now:
+- 16:9 card and square card (attached)
+- 20-second demo clip (attached): a GLB dropped into Rig Doctor, the unmapped joints listed, then the mapping and test that fix it
+- Alt text for each, and a one-line description: "Give your AI a body. Any rig, any page, one tag."
+
+If it helps the social post, the clip is the part that shows the live contribution. Anything else you need from us for the prep doc or the StreamYard link, we can turn around the same day.
+```
+
+Addressed to the intake GitHub named: the issue thread itself. No host is tagged by name beyond the
+assignees GitHub set on the issue.
+
+## Event package
+
+`{{EVENT_TIME_UTC}}` is the booked slot written as `YYYY-MM-DD HH:MM UTC`. The show airs Fridays at
+1:00 PM ET, which is 17:00 UTC until US daylight time ends on 2026-11-01 and 18:00 UTC after. It is
+filled when the booking confirms; nothing below ships with the token still in it.
+
+**Event page description** (for the host's prep doc and any listing that asks for one):
+
+> Give your AI a body. Any rig, any page, one tag. three.ws is an Apache-2.0 framework for 3D AI
+> agents on the web. On this stream we generate a rigged avatar, embed it on an empty page with one
+> `<agent-3d>` tag, then take a skeleton convention the community proposed, map it, test it, and open
+> the pull request live. Watch at https://www.twitch.tv/github.
+
+**T-7 demo clip, 20 seconds, 1920x1080, browser zoom 125%:**
+
+| Time | Shot | Route |
+|---|---|---|
+| 0:00 to 0:04 | Drop a humanoid GLB onto Rig Doctor; the detected convention name appears | https://three.ws/rig-doctor |
+| 0:04 to 0:09 | Scroll the joint table to the rows that will not animate | same |
+| 0:09 to 0:14 | Editor split: the new mapping entry in `src/glb-canonicalize.js`, then `npx vitest run tests/glb-canonicalize.test.js` passing | repo |
+| 0:14 to 0:18 | Re-run the same GLB in Rig Doctor; the joints now map | https://three.ws/rig-doctor |
+| 0:18 to 0:20 | End card: "Open Source Friday, {{EVENT_TIME_UTC}}, twitch.tv/github" | card |
+
+Record the 0:09 to 0:18 segment on a real proposed convention once one is chosen; do not stage a fake
+failure.
+
+**T-1 reminder** (176 weighted characters with the token in place):
+
+```text
+Tomorrow at {{EVENT_TIME_UTC}}, on GitHub's Open Source Friday: a new rig convention mapped, tested, and opened as a PR live on stream. Watch on Twitch: https://www.twitch.tv/github
+```
+
+Add one converted local time in the image, not the text, per plan section 5.
+
+**Live run of show:** the 45-minute runsheet in
+[plan section 4](../../../docs/open-source-friday-plan.md#4-stream-runsheet-30-to-45-min-live-at-100-pm-et-join-1245-for-tech-check)
+is the run of show; it follows GitHub's own Streaming Guide. Swap its #110/#111 references for the
+chosen proposal's new issue number.
+
+**T+1 recap skeleton** (fill every field from its source; if a source cannot supply it, write
+"not available" rather than a number):
+
+```text
+Open Source Friday recap: {{PR_TITLE}} went from proposal to merged on air.
+
+Replay: {{REPLAY_URL}}
+The pull request: {{PR_URL}} by {{CONTRIBUTOR_HANDLE}}
+Conventions Rig Doctor recognises now: {{CONVENTION_COUNT}}
+Issues claimed during the week: {{ISSUES_CLAIMED}}
+Next: {{NEXT_ISSUE_URL}}
+```
+
+| Field | Source |
+|---|---|
+| `{{REPLAY_URL}}` | GitHub's YouTube playlist linked from the githubevents/open-source-friday README |
+| `{{PR_URL}}`, `{{PR_TITLE}}`, `{{CONTRIBUTOR_HANDLE}}` | `gh pr list -R nirholas/three.ws --state merged --limit 5` |
+| `{{CONVENTION_COUNT}}` | `CONVENTION_COUNT` exported by `src/rig-report.js` on `main` |
+| `{{ISSUES_CLAIMED}}` | `gh issue list -R nirholas/three.ws --label "good first issue" --state all` with an assignee or claim comment since T-14 |
+| `{{NEXT_ISSUE_URL}}` | the next open `good first issue` |
+| Landing sessions (scorecard, not the post) | web analytics, `utm_campaign=mkt-2026-09-osf` |
+
+## Claims check
+
+| Claim | Proof | Verified |
+|---|---|---|
+| Approved for Open Source Friday | `approved` label on issue #254 (`gh issue view 254 -R githubevents/open-source-friday`) | 2026-09-17 |
+| No date booked yet | Same issue: one comment, GitHub's booking request | 2026-09-17 |
+| Rig Doctor recognises 15 conventions, including MikuMikuDance, Kinect, 3ds Max Biped | `CONVENTIONS` in `src/rig-report.js` (15 entries) | 2026-09-17 |
+| The GLB never leaves the tab | comment in `src/rig-doctor-page.js` line 14 ("never leaves the tab") | 2026-09-17 |
+| All six curated starter issues are closed | `gh issue view 110` through `115`, state CLOSED | 2026-09-17 |
+| A convention is a mapping entry plus a test | `docs/first-contribution.md` worked example; CLAUDE.md animation rule | 2026-09-17 |
+| Clone to PR in about 15 minutes | `docs/first-contribution.md` | repo |
+| Airs Fridays 1 PM ET on twitch.tv/github | githubevents/open-source-friday README | 2026-09-17 |
+| Discussions Ideas category exists | GraphQL `discussionCategories` lists Ideas; URL returns 200 | 2026-09-17 |
+
+Dropped: star count (moves daily), "merged on air" as a promise in the anchor (depends on review on the
+day; the anchor says the PR is opened live).
+
+## Owner does
+
+Book a Friday at `https://gh.io/osf-booking` and post the booked date on issue #254; then fill
+`{{EVENT_TIME_UTC}}` and publish the X anchor with the Rig Doctor image.
diff --git a/marketing/growth/kits/MKT-2026-10-NVIDIA-DH.md b/marketing/growth/kits/MKT-2026-10-NVIDIA-DH.md
new file mode 100644
index 000000000..00864b9aa
--- /dev/null
+++ b/marketing/growth/kits/MKT-2026-10-NVIDIA-DH.md
@@ -0,0 +1,140 @@
+# MKT-2026-10-NVIDIA-DH: audio drives a visitor-generated 3D face in the browser
+
+**Kit status:** copy complete. **Blocked: the proof page is broken in production.** On 2026-09-17
+`https://three.ws/demos/audio2face` (and `/audio2face`) served its inline module unbundled: it imports
+the bare specifier `'three'`, the browser throws `Failed to resolve module specifier "three"`, and the
+avatar never leaves "Loading avatar...". The API underneath works (see claims check). Nothing in this
+kit posts until the page renders a speaking avatar in a clean browser.
+
+## At a glance
+
+| Field | Value |
+|---|---|
+| Campaign id | `MKT-2026-10-NVIDIA-DH` |
+| Publish date | 2026-10-06 |
+| Summary | A typed line is spoken by NVIDIA Magpie, NVIDIA Audio2Face-3D turns that audio into a blendshape track, and the browser maps it onto whatever face the visitor's avatar has |
+| Audience | GPU AI and 3D developers |
+| Primary channel | NVIDIA Developer Forums |
+| Secondary channels | X, LinkedIn, Telegram community, three.ws news |
+| Proof | https://three.ws/demos/audio2face |
+| Tracked links | X `https://three.ws/demos/audio2face?utm_source=x&utm_medium=social&utm_campaign=mkt-2026-10-nvidia-dh&utm_content=anchor`; LinkedIn `...utm_source=linkedin&utm_medium=social&...&utm_content=post`; Telegram `...utm_source=telegram&utm_medium=community&...&utm_content=announce`; forum `...utm_source=nvidia-forums&utm_medium=community&...&utm_content=forum-post` |
+| The single CTA | Run the demo and inspect the implementation |
+| Partner ask | Inception reshare and Startup Showcase consideration |
+| KPI | demo starts and partner amplification |
+
+## Audit: what already exists
+
+| Channel | Existing asset | State | How this kit uses it |
+|---|---|---|---|
+| NVIDIA Developer Forums (primary) | [docs/nvidia-forum-browser-digital-human.md](../../../docs/nvidia-forum-browser-digital-human.md) | Draft, unposted | **Is** the primary post. Its latencies are from 2026-09-02; re-measure on posting day |
+| X, Audio2Face clip post | [marketing/nvidia-inception/social-copy.md](../../nvidia-inception/social-copy.md) "The Audio2Face clip" | Written, over the 179 target | Replaced by the tighter anchor below; its rules (Inception member wording, `#NVIDIAInception`, `@NVIDIAAIDev`, no coin) govern this kit |
+| LinkedIn and Telegram | same file | Membership beat, not the digital-human beat | Not reused for this campaign |
+| Partner email | [templates.md, NVIDIA partner amplification note](../templates.md) | Ready | Reused with the two fills below |
+| Intake | [docs/nvidia-visibility-map.md](../../../docs/nvidia-visibility-map.md): existing `[email protected]` thread | Current | The only address used |
+| Media | [marketing/nvidia-inception/assets/three-ws-audio2face-1920x1080.png](../../nvidia-inception/assets/three-ws-audio2face-1920x1080.png) | Exists | Anchor image, after the page is fixed and the frame re-checked |
+| X anchor, LinkedIn and Telegram for this beat, news | none | Missing | Written below |
+
+## X
+
+**Media:** the Audio2Face asset above, or better, a 15-second screen capture of the fixed page with
+browser chrome visible. **Alt text:** The three.ws Audio2Face demo: a 3D avatar speaking a typed line,
+with live blendshape meters for jaw and mouth moving beside it.
+
+**Anchor** (175 weighted characters):
+
+```text
+Type a line and NVIDIA Audio2Face-3D returns 55 blendshapes at 30 fps, mapped onto the face your avatar already has. In a browser tab: https://three.ws/demos/audio2face?utm_source=x&utm_medium=social&utm_campaign=mkt-2026-10-nvidia-dh&utm_content=anchor #NVIDIAInception
+```
+
+The one hashtag is a deliberate exception to the voice contract's zero-hashtag rule: the NVIDIA kit
+requires `#NVIDIAInception` because it is the tag NVIDIA's social team monitors, and a reshare is this
+campaign's partner ask.
+
+**Thread:**
+
+2/ (180)
+```text
+The loop: Magpie TTS speaks the line, Audio2Face-3D turns that exact audio into a per-frame track, and the browser samples the track against the playing audio's clock. @NVIDIAAIDev
+```
+
+3/ (178)
+```text
+The hard part is the rig. A2F speaks ARKit; many avatars only ship VRM vowels or Oculus visemes. We rebuild each vowel from the ARKit frame, so a five-vowel face still lip-syncs.
+```
+
+4/ (123 with the token)
+```text
+The full write-up, with the resampling trap and measured round trips, is on the NVIDIA Developer Forums: {{FORUM_POST_URL}}
+```
+
+## LinkedIn
+
+```text
+Audio2Face-3D is usually shown inside a game engine, driving a character somebody built on purpose. We run it in a browser tab, on a character the visitor generated themselves.
+
+Type a line on the three.ws demo. NVIDIA Magpie speaks it, and NVIDIA Audio2Face-3D turns that exact audio into a per-frame track of 55 blendshape weights at 30 frames per second. The browser plays the original audio and samples the track against its clock, so the lips follow the real voice.
+
+The difficult part was never the model. It was the rig: we do not choose the avatar's face. Some ship ARKit shapes, some only VRM vowels or Oculus visemes. Where a face has only vowels, we reconstruct each one from the ARKit frame, so it still lip-syncs.
+
+We wrote up the implementation, including the resampling trap and measured latencies, on the NVIDIA Developer Forums, and the code is open source.
+
+three.ws is a member of NVIDIA Inception. NVIDIA does not endorse three.ws.
+
+Run it: https://three.ws/demos/audio2face?utm_source=linkedin&utm_medium=social&utm_campaign=mkt-2026-10-nvidia-dh&utm_content=post
+
+#NVIDIAInception
+```
+
+(172 words.) Tag **NVIDIA for Startups** on LinkedIn, per the NVIDIA kit.
+
+## Telegram community
+
+```text
+A talking 3D face, in a browser tab, on NVIDIA models.
+
+Type a line: NVIDIA Magpie speaks it and NVIDIA Audio2Face-3D animates the avatar's face from that exact audio, 30 frames a second. It adapts to whatever face the avatar has, including rigs that only know a few vowel shapes.
+
+Try it, then read how it works on the NVIDIA Developer Forums.
+https://three.ws/demos/audio2face?utm_source=telegram&utm_medium=community&utm_campaign=mkt-2026-10-nvidia-dh&utm_content=announce
+```
+
+No coin in this post or thread, per the NVIDIA kit.
+
+## three.ws news blurb
+
+The three.ws Audio2Face demo drives a 3D avatar's face in the browser from NVIDIA Audio2Face-3D: a
+typed line is spoken by NVIDIA Magpie, and the returned blendshape track animates the face frame by
+frame. It maps onto ARKit, VRM, and Oculus-style rigs without a game engine, and the full implementation
+is written up on the NVIDIA Developer Forums: {{FORUM_POST_URL}}.
+
+## Partner message
+
+Use the existing [NVIDIA partner amplification note](../templates.md) in the existing
+`[email protected]` thread, with two changes:
+
+1. Replace the demo URL line with the forum post: `We published the implementation on the NVIDIA Developer Forums: {{FORUM_POST_URL}}. The live demo is https://three.ws/demos/audio2face.`
+2. Append one sentence: `If the story fits, we would like three.ws considered for the Inception Startup Showcase.`
+
+The visibility map says NVIDIA did not answer the 2026-09-04 Showcase routing question; this reply
+continues that thread and does not open a new one.
+
+## Claims check
+
+| Claim | Proof | Verified |
+|---|---|---|
+| Audio2Face-3D returns 55 blendshape weights at 30 fps | live `POST https://three.ws/api/a2f` with `{"text":"Hello from a browser tab."}`: `fps` 30, `blendShapeNames` length 55, `model` `audio2face-3d` | 2026-09-17 |
+| Magpie speaks the line first, one call | same response carries `audio` and `animation`; handler comment in `api/a2f.js` ("One-shot text to speech to animation, server synthesizes via Magpie") | 2026-09-17 |
+| Track sampled against the audio clock | [forum draft](../../../docs/nvidia-forum-browser-digital-human.md), "The audio contract"; `src/voice/a2f-player.js` | repo |
+| Maps onto ARKit, VRM vowels, Oculus visemes; vowels rebuilt from ARKit | `src/voice/arkit-blendshapes.js`, `src/voice/a2f-player.js` header comment | 2026-09-17 |
+| Runs in a browser tab, no engine | **Currently false in production**: the page fails to load `three`. Restore before posting | 2026-09-17 |
+| Member of NVIDIA Inception; not a partner, no endorsement | [docs/partners.md](../../../docs/partners.md), [marketing/nvidia-inception/social-copy.md](../../nvidia-inception/social-copy.md) | repo |
+| Code is open source | Apache-2.0 per `gh repo view` | 2026-09-17 |
+
+Dropped: the round-trip latencies (measured 2026-09-02 in the draft, not re-measured here; the forum
+post carries them after re-measurement) and "a character the visitor generated ninety seconds earlier"
+(generation time not re-measured).
+
+## Owner does
+
+Get `/demos/audio2face` rendering again, then post the forum draft and publish the X anchor with the
+demo capture.
diff --git a/marketing/growth/kits/images/huggingface-org.png b/marketing/growth/kits/images/huggingface-org.png
new file mode 100644
index 000000000..2d791568d
Binary files /dev/null and b/marketing/growth/kits/images/huggingface-org.png differ
diff --git a/marketing/growth/submissions/awesome-web-components.md b/marketing/growth/submissions/awesome-web-components.md
new file mode 100644
index 000000000..e758e695f
--- /dev/null
+++ b/marketing/growth/submissions/awesome-web-components.md
@@ -0,0 +1,92 @@
+# Awesome Web Components submission
+
+Target: [web-padawan/awesome-web-components](https://github.com/web-padawan/awesome-web-components)
+(3,580 stars, last push 2026-09-12, most recent merged additions 2026-08-17, 2026-08-27 and
+2026-09-09).
+
+Contribution path: fork, edit `README.md`, open one pull request. Rules live in
+[CONTRIBUTING.md](https://github.com/web-padawan/awesome-web-components/blob/main/CONTRIBUTING.md).
+
+Date verified: **2026-09-17**.
+
+## Product
+
+One product only: the `<agent-3d>` custom element, shipped as
+[`@three-ws/avatar`](https://github.com/nirholas/three.ws/tree/main/avatar-sdk) and as a
+script-tag loader at `https://three.ws/agent-3d/latest/agent-3d.js`.
+
+## Why it fits
+
+The list's **Real World > Components** section catalogues standalone, framework-agnostic
+custom elements that do one job, including `<model-viewer>` for static 3D models. `<agent-3d>`
+is the same shape of thing for animated humanoid avatars: one tag, a GLB URL, no build step, and
+opt-in chat, voice, and lipsync. Web component builders who reach `<model-viewer>` and need a
+character that moves are exactly the reader this entry serves.
+
+## Section
+
+`## Real World` > `### Components`, alphabetical by tag name: directly after
+``[`<active-table>`]`` and before ``[`<api-viewer>`]``.
+
+## Listing line
+
+```markdown
+- [`<agent-3d>`](https://github.com/nirholas/three.ws/tree/main/avatar-sdk) - Web component that renders an animated 3D avatar from a GLB, with optional chat, voice, and lipsync.
+```
+
+## Pull request title
+
+```text
+Add <agent-3d> to Components
+```
+
+## Pull request body
+
+```markdown
+Adds `<agent-3d>` to Real World > Components, in alphabetical order.
+
+`<agent-3d>` is a custom element for animated 3D avatars. A bare
+`<agent-3d body="https://example.com/avatar.glb">` renders an animated character with no API key
+and no framework; chat, voice input and output, and audio-driven lipsync are opt-in attributes.
+It sits next to `<model-viewer>` in purpose: model-viewer shows an object, agent-3d shows a
+character that moves.
+
+- Source: https://github.com/nirholas/three.ws/tree/main/avatar-sdk (Apache-2.0)
+- npm: `npm install @three-ws/avatar three`
+- API reference: https://three.ws/docs/web-component
+
+Minimal usage:
+
+ <script type="module" src="https://three.ws/agent-3d/latest/agent-3d.js"></script>
+ <agent-3d body="https://three.ws/avatars/default.glb" style="width:400px;height:500px"></agent-3d>
+```
+
+## Evidence (checked 2026-09-17)
+
+| Claim | Check | Result |
+| --- | --- | --- |
+| Not already listed | README fetched via GitHub API, searched for `three.ws`, `nirholas`, `threews`, `agent-3d` | 0 matches |
+| No duplicate suggestion | `gh search prs` / `gh search issues` in the repo for `three.ws` and `nirholas` | 0 results |
+| Source link is live | `curl -L` on the `avatar-sdk` tree URL | 200 |
+| Loader is live | `curl` on `https://three.ws/agent-3d/latest/agent-3d.js` | 200 |
+| API reference is live | `curl` on `https://three.ws/docs/web-component` | 200 |
+| Package exists | `npm view @three-ws/avatar` | 0.2.3, Apache-2.0, `repository.directory: avatar-sdk` |
+| Install works | `npm install @three-ws/avatar three` in a clean scratch project | succeeded |
+| The minimal snippet renders with no key | Headless Chromium loaded the exact snippet above | `agent:ready` fired, no page errors, avatar visible in the screenshot |
+
+The package's main entry references `window`, so it is browser-only; importing it in plain Node
+throws. That is expected for a custom element and matches how every other entry in the section
+is consumed.
+
+## Target rules complied with
+
+- Format `[Name](Link) - Brief description`, matching the tag-name style the section already uses.
+- Description does not start with "A" or "An".
+- Added to the relevant category in alphabetical order.
+- One addition, useful PR title.
+- Searched previous suggestions first (none).
+
+## Owner's single action
+
+Fork the repo, paste the listing line after `<active-table>` in Components, and open the pull
+request with the title and body above.
diff --git a/marketing/growth/submissions/awesome-webxr.md b/marketing/growth/submissions/awesome-webxr.md
new file mode 100644
index 000000000..82522b745
--- /dev/null
+++ b/marketing/growth/submissions/awesome-webxr.md
@@ -0,0 +1,90 @@
+# Awesome WebXR submission
+
+Target: [msub2/awesome-webxr](https://github.com/msub2/awesome-webxr) (283 stars, last push
+2026-08-21, one open PR, last merged additions 2026-03-07 and 2025-08-29).
+
+Contribution path: fork, edit `README.md`, open a pull request. Rules live in
+[contributing.md](https://github.com/msub2/awesome-webxr/blob/main/contributing.md).
+
+Date verified: **2026-09-17**.
+
+## Product
+
+One product only: **3D AR Studio**, the standalone open-source `<ar-studio>` element
+([nirholas/3D-AR-Studio](https://github.com/nirholas/3D-AR-Studio), npm
+[`3d-ar-studio`](https://www.npmjs.com/package/3d-ar-studio), live demo
+<https://nirholas.github.io/3D-AR-Studio/>). It is the extracted, generalized version of the
+three.ws `/ar/studio` surface.
+
+## Why it fits
+
+The list is small and hand-curated for people building WebXR on the web. Its
+**Development > Frameworks and Libraries** section lists drop-in libraries (A-Frame, three.js,
+Threlte). 3D AR Studio is a library a site adds with one script tag, and its core is real WebXR
+work readers of this list care about: `immersive-ar` sessions, one `XRAnchor` per placed model,
+hit-test reticle, light estimation, and depth occlusion, with Quick Look and Scene Viewer as the
+honest fallbacks where WebXR does not exist.
+
+## Section
+
+`## Development` > `### Frameworks and Libraries`. The section is alphabetical and entries are
+separated by a blank line, so the entry goes first, above `A-Frame`.
+
+## Listing line
+
+```markdown
+- [3D AR Studio](https://nirholas.github.io/3D-AR-Studio/) - A drop-in `<ar-studio>` web component that places many 3D models in one live camera view using WebXR anchors, hit testing, light estimation, and depth occlusion, with Quick Look and Scene Viewer handoff where WebXR is unavailable. [[repo]](https://github.com/nirholas/3D-AR-Studio)
+```
+
+## Pull request title
+
+```text
+Add 3D AR Studio to Frameworks and Libraries
+```
+
+## Pull request body
+
+```markdown
+Adds 3D AR Studio under Development > Frameworks and Libraries.
+
+3D AR Studio is an Apache-2.0 web component for multi-model AR on the web. One script tag gives
+a page a camera view where people can place, move, pinch-resize, and rotate any number of GLB
+models. On devices with WebXR it runs an `immersive-ar` session with one XRAnchor per model,
+light estimation, and depth occlusion; on iOS it hands off to Quick Look and on Android without
+WebXR to Scene Viewer. Scenes round-trip through the URL, so a layout composed on desktop
+reopens on a phone from a QR code.
+
+- Demo: https://nirholas.github.io/3D-AR-Studio/
+- Repo: https://github.com/nirholas/3D-AR-Studio
+- npm: https://www.npmjs.com/package/3d-ar-studio
+
+ <script type="module" src="https://unpkg.com/3d-ar-studio/dist/ar-studio.min.js"></script>
+ <ar-studio></ar-studio>
+```
+
+## Evidence (checked 2026-09-17)
+
+| Claim | Check | Result |
+| --- | --- | --- |
+| Not already listed | README fetched via GitHub API, searched for `three.ws`, `nirholas`, `threews`, `agent-3d`, `AR Studio` | 0 matches |
+| No duplicate suggestion | `gh search prs` / `gh search issues` for `three.ws` and `nirholas` | 0 results |
+| Demo is live | `curl -L https://nirholas.github.io/3D-AR-Studio/` | 200 |
+| Repo is public | `gh api repos/nirholas/3D-AR-Studio` | public, Apache-2.0, created 2026-08-19, last push 2026-09-15 |
+| Package exists | `npm view 3d-ar-studio` | 0.3.1, Apache-2.0, bins `3d-ar-studio`, `ar-studio`, `3d-ar-studio-mcp` |
+| Install works | `npm install 3d-ar-studio` in a clean scratch project | succeeded |
+| WebXR anchors, light estimation, depth occlusion, Quick Look, Scene Viewer | Package README ("Why this exists") and the three.ws implementation it was extracted from (`src/ar/multi-place.js`, `src/ar/estimated-lighting.js`, `src/ar/depth-occlusion.js`) | present |
+
+Device AR itself was not exercised on hardware in this pass; the claims above are the shipped
+code paths, not a device test.
+
+## Target rules complied with
+
+- Style `[Site](URL) - Description`, matching the neighbouring entries, with the `[[repo]]`
+ suffix other entries (XRSH, XR Fragments) use for a separate source link.
+- Placed in the most specific existing category, alphabetical position.
+- No new category or restructuring.
+
+## Owner's single action
+
+Before opening the PR, load the demo on one Android phone and one iPhone and confirm a model
+places. Then fork, paste the entry above `A-Frame`, and open the pull request.
diff --git a/marketing/growth/submissions/khronos-gltf-project-explorer.md b/marketing/growth/submissions/khronos-gltf-project-explorer.md
new file mode 100644
index 000000000..1bdddf458
--- /dev/null
+++ b/marketing/growth/submissions/khronos-gltf-project-explorer.md
@@ -0,0 +1,89 @@
+# Khronos glTF Project Explorer submission
+
+Target: the Khronos Group's official glTF ecosystem registry, the
+[glTF Project Explorer](https://github.khronos.org/glTF-Project-Explorer/), now served through
+the [Khronos Ecosystem Explorer](https://ecosystem.khronos.org/khronos-standard/gltf/).
+
+Contribution path: the web form at <https://ecosystem.khronos.org/submit/>. The
+[explorer repository README](https://github.com/KhronosGroup/glTF-Project-Explorer) says: "To
+submit new projects, go to: https://ecosystem.khronos.org/submit/". The repository itself takes
+no content PRs (its last merges were dependency bumps in 2024-05), so do not open a PR there.
+
+Date verified: **2026-09-17**.
+
+## Product
+
+One product only: **three.ws Model Diff**, the structural glTF/GLB diff published as
+[`@three-ws/glb-diff`](https://github.com/nirholas/three.ws/tree/main/packages/glb-diff) (library
+plus `glb-diff` CLI), with the browser version at <https://three.ws/diff>.
+
+## Why it fits
+
+The explorer is the first stop for people looking for glTF tooling by task. Its data already
+has `analyze`, `inspect`, and `validate` tasks, and 329 projects, but nothing that answers "what
+changed between two versions of this model". A diff that reports geometry, material, texture,
+skeleton, and animation changes with a severity a CI job can gate on is a genuine gap in the
+glTF pipeline, and it is built on glTF-Transform, so it reads glTF the way the ecosystem does.
+
+## Category
+
+Map to the explorer's own data vocabulary (read from the live dataset
+`public/data/glTF-projects-data.json`):
+
+| Field | Value |
+| --- | --- |
+| task | `analyze`, `inspect`, `validate` |
+| type | `library`, `application`, `web application` |
+| language | `JavaScript` |
+| license | `Apache-2.0` |
+| inputs | `glTF 2.0`, `GLB` |
+| outputs | none (it produces text, Markdown, and JSON reports, not models) |
+
+## Form answers
+
+The submission form sits behind a Cloudflare Turnstile challenge, which returned 403 to curl,
+WebFetch, and headless Chromium on 2026-09-17, so its exact field labels could not be read. The
+answers below cover every field the published dataset stores for a project; fill them into
+whichever labels the form shows.
+
+```text
+Name: three.ws Model Diff (glb-diff)
+Link: https://github.com/nirholas/three.ws/tree/main/packages/glb-diff
+Description: Structural diff for glTF 2.0 and GLB files. Reports what changed between two
+ versions of a model (geometry, materials, textures, skeletons, animation clips)
+ with git-style rename and move detection, and assigns a severity (none, cosmetic,
+ minor, major, breaking) that a CI job can fail on. Ships as a JavaScript library,
+ a CLI with text, Markdown, and JSON output, and a browser tool at
+ https://three.ws/diff that runs locally without uploading the models.
+Tasks: analyze, inspect, validate
+Type: library, application, web application
+Language: JavaScript
+License: Apache-2.0
+Inputs: glTF 2.0, GLB
+Outputs: (leave empty)
+```
+
+## Evidence (checked 2026-09-17)
+
+| Claim | Check | Result |
+| --- | --- | --- |
+| Not already listed | Live dataset fetched from the repo and searched for `three.ws`, `nirholas`, `glb-diff` | 0 matches |
+| No prior request | Issue list of the explorer repo | no three.ws issue |
+| Source link is live | `curl -L` on the `packages/glb-diff` tree URL | 200 |
+| Browser tool is live | `curl -L https://three.ws/diff` | 200 |
+| Package exists | `npm view @three-ws/glb-diff` | 0.1.0, Apache-2.0, bin `glb-diff` |
+| Install and CLI work | `npm install @three-ws/glb-diff`, then `npx glb-diff a.glb b.glb` on two copies of `https://three.ws/avatars/default.glb` | "identical: every scene, mesh, material, texture, skeleton and clip matches.", exit 0 |
+| Reads JSON glTF as well as GLB | `src/document.js` checks the GLB magic and otherwise parses JSON glTF | present |
+| Browser tool does not upload | `/diff` page description in `data/pages.json`: "Runs in your browser, nothing is uploaded." | present |
+
+## Target rules complied with
+
+- Submitted through the channel the explorer README names, not a repo PR.
+- Values use the dataset's existing vocabulary, so the entry filters correctly.
+- One project per submission.
+
+## Owner's single action
+
+Open <https://ecosystem.khronos.org/submit/> in a normal browser, pass the challenge, and paste
+the form answers above.