BluePrint AI

AI-powered 3D car explorer. Launch page, interactive garage, Featherless assistant, Supabase verified specs, Flutter reference client.

Live: https://blueprint-ai-kohl.vercel.app
Repo: https://github.com/ShadowEsu/Blueprint

What’s in this repo

Piece	Path	Purpose
Launch site	`/` (`index.html`, `src/`)	Marketing scroll page + 3D hero car
Final Product	`/app.html` (`src/app.js`)	Main app — 8 machines, explode, paint, AI assistant
Explorer	`/explorer/`	Alternate garage / anatomy / wind tunnel
AI agent	`/api/agent` (`api/agent.js`)	Featherless LLM + Supabase data tools
Health check	`/api/health`	Supabase + provider status
Supabase	`supabase/`	Postgres schema, migrations, seed data
Flutter	`integrations/blueprint-flutter/`	Mobile reference client (Supabase + AI chat)
Local agent	`scripts/serve-agent.mjs`	Dev server for AI on port 8000

Stack: Static site on Vercel · Three.js 3D · Featherless AI · Supabase Postgres · Flutter (reference)

Architecture (what we actually run)

These diagrams match the real files and services in this repo — not mockups.

System overview

flowchart TB
  subgraph Browser["Browser"]
    Launch["/ index.html + src/main.js"]
    Explorer["/explorer/ DC + Three.js"]
    Bridge["agent-bridge.js"]
  end

  subgraph Vercel["Vercel — blueprint-ai"]
    Static["Static files\nindex, explorer, api"]
    AgentFn["/api/agent\napi/agent.js"]
  end

  subgraph External["External services"]
    Featherless["Featherless API\nLLM + tool calls"]
    Supabase["Supabase Postgres\nschema + RLS"]
    jsDelivr["jsDelivr CDN\nGitHub main branch"]
  end

  subgraph LocalDev["Local dev optional"]
    PyServer["python http.server :4173"]
    AgentDev["serve-agent.mjs :8000"]
  end

  subgraph FlutterApp["Flutter reference client"]
    Flutter["integrations/blueprint-flutter"]
  end

  Launch --> Explorer
  Explorer --> Bridge
  Bridge -->|POST production| AgentFn
  Bridge -->|POST localhost| AgentDev
  AgentFn --> Featherless
  AgentFn --> Supabase
  AgentDev --> Featherless
  AgentDev --> Supabase
  Explorer -->|resolveAsset on *.vercel.app| jsDelivr
  Explorer -->|local paths| Static
  Flutter --> Supabase
  PyServer -.->|npm run dev| Launch

User journey (pages)

flowchart LR
  A["/ Launch site\nThree.js hero car"] -->|Launch BluePrint| B["Garage.dc.html\n4 cars + wind tunnel"]
  B -->|car=0| C["Car Anatomy Explorer\nGT3 RS 114 parts"]
  B -->|car=1-3| D["Car Anatomy Explorer\nmonolithic GLB"]
  B --> E["Wind Tunnel.html\naero viz"]
  C -->|Engine chip| F["engine_teardown/engine.html\nflat-6 layers"]
  C -->|Gold orb| G["AI assistant\npaint / explode / specs"]

AI assistant request flow

sequenceDiagram
  participant U as User
  participant UI as Car Anatomy Explorer
  participant B as agent-bridge.js
  participant A as api/agent.js
  participant F as Featherless
  participant S as Supabase
  participant T as Three.js scene

  U->>UI: "paint it guards red"
  UI->>B: BlueprintAgent.ask()
  B->>A: POST /api/agent
  A->>F: chat.completions + tools
  F-->>A: set_paint, speech
  A->>S: get_part_facts optional
  S-->>A: verified specs
  A-->>B: speech + actions[]
  B-->>UI: JSON response
  UI->>T: runChatActions set_paint
  T-->>U: body meshes recolored
  Note over B,A: Offline keyword fallback if API down

GT3 RS multipart load

flowchart TD
  M["partsManifest\n114 part names in Car Anatomy Explorer.dc.html"] --> R["resolveAsset\nasset-config.js"]
  R -->|localhost| L["./explorer/assets/gt3rs/parts/*.glb\ndisk"]
  R -->|*.vercel.app| J["cdn.jsdelivr.net/gh/ShadowEsu/Blueprint@main/explorer/assets/gt3rs/parts/"]
  L --> GL["GLTFLoader × 114"]
  J --> GL
  GL --> C["THREE.Group combined"]
  C --> P["processCar\ncategorize → wing hood doors wheels engine…"]
  P --> V["Orbit + explode + AI paint targets"]

Asset deployment split

flowchart LR
  GH["GitHub main\nShadowEsu/Blueprint"] --> V["Vercel deploy\n.vercelignore strips large GLBs"]
  GH --> JD["jsDelivr\nserves excluded GLBs"]
  V --> Small["Shipped on Vercel\nHTML JS CSS logo\nshowroom-car.glb ~33MB"]
  JD --> Large["CDN only\nexplorer/assets/*.glb\ngt3rs/parts/*.glb ×114"]
  Small --> User["Browser"]
  Large --> User

Supabase data layer

erDiagram
  cars ||--o{ car_variants : has
  car_variants ||--o| car_specs : has
  car_variants ||--o{ performance_tests : has
  car_variants ||--o{ pricing_market : has
  cars ||--o{ research_chunks : references

  cars {
    text id PK
    text make
    text model
    text summary
  }
  car_variants {
    text id PK
    text car_id FK
    text drivetrain
    text transmission
  }
  car_specs {
    text variant_id FK
    text engine
    int horsepower_hp
  }
  research_chunks {
    text chunk_title
    text chunk_text
  }

Agent reads these tables in api/agent.js (service role server-side). Flutter reads with anon key + RLS public read.

Tech stack map

mindmap
  root((BluePrint AI))
    Frontend
      index.html launch
      src/main.js Three.js
      explorer DC runtime
      support.js React
      Three.js GLTFLoader
    Hosting
      Vercel static
      Vercel serverless api
      jsDelivr GLB CDN
    AI
      Featherless Qwen2.5-7B
      api/agent.js tools
      agent-bridge.js client
    Data
      Supabase Postgres
      seed fallback in agent
    Mobile
      Flutter reference app
      supabase_flutter

Quick start (local)

# 1. Install deps
npm install

# 2. Env (copy template, fill keys — never commit .env)
cp .env.example .env

# 3. Static site
npm run dev
# → http://localhost:4173

# 4. AI agent (separate terminal — required for full assistant)
npm run dev:agent
# → http://127.0.0.1:8000

Open http://localhost:4173 → Launch BluePrint → Garage.

Do not double-click index.html. ES modules need a server (npm run dev or Vercel).

Live routes

Page	URL
Launch site	`/`
Final Product (main)	`/app.html`
Health / Supabase check	`/api/health`
Garage (explorer)	`/explorer/Garage.dc.html`
Car anatomy (3D)	`/explorer/Car%20Anatomy%20Explorer.dc.html`
GT3 by index	`/explorer/Car%20Anatomy%20Explorer.dc.html?car=0`
Wind tunnel	`/explorer/Wind%20Tunnel.html`
GT3 engine teardown	`/explorer/uploads/engine_teardown/engine.html?car=gt3rs`
AI API	`POST /api/agent`

Launch site (`/`)

Scroll sections: hero, problem, AI stack (Featherless / FlutterAI / Supabase), launch CTA
Launch BluePrint and Launch App → /explorer/Garage.dc.html
Hero 3D car: assets/blueprint-showroom-car.glb (deployed on Vercel; mobile loads a fast fallback first)
Styles: src/styles.css · Scene: src/main.js (Three.js via import map)

Explorer (`/explorer/`)

Interactive Car Assembly Explorer 2 — the main product.

Garage

Four cars, each opens the anatomy viewer:

Index	Car	Notes
`0`	Porsche 911 GT3 RS	114-part multipart assembly (`assets/gt3rs/parts/`)
`1`	Ferrari LaFerrari	Monolithic GLB
`2`	Porsche 911 Turbo	Monolithic GLB
`3`	Lamborghini Aventador SVJ	Monolithic GLB

Also links to Wind Tunnel aerodynamics view.

Car Anatomy Explorer

Orbit, zoom, click parts, explode slider, category chips (wing, hood, doors, wheels, engine, exhaust, interior)
GT3 RS engine chip → engine teardown page (not generic engine select)
AI orb (bottom right) → assistant chat

Runtime: support.js (DC/React) + Three.js import map + asset-config.js for CDN paths.

GT3 RS model

Uses partsManifest + partsBase, not the old single porsche_gt3_rs.glb:

explorer/assets/gt3rs/parts/*.glb   # 114 segmented parts

Part names are categorized in Car Anatomy Explorer.dc.html (categorize() for wing, hood, doors, wheels, engine, etc.).

On Vercel, these parts load from jsDelivr (GitHub main branch) because they exceed deploy size limits. First load can take 30–60s while the CDN warms up. Locally they load from disk instantly.

Wind Tunnel

Illustrative airflow over four monolithic GLB models (explorer/assets/*.glb).

Engine teardown

explorer/uploads/engine_teardown/engine.html?car=gt3rs — layered flat-6 teardown with explode slider.

AI assistant

How it works

User types in the gold orb chat (Car Anatomy Explorer)
explorer/agent-bridge.js sends POST to:
- Production: /api/agent
- Local: http://127.0.0.1:8000/ (run npm run dev:agent)
api/agent.js calls Featherless (OpenAI-compatible) with tool calling
Supabase (or in-memory seed) backs factual answers
Client runs scene actions on the live Three.js model

If the API is unreachable, a keyword offline fallback still handles paint / explode / reset / show engine.

Scene tools (client executes these)

Tool	What it does
`set_paint`	Change body/hood/door paint (name or hex)
`focus_camera`	Frame a part or angle (front, rear, engine, …)
`highlight`	Cyan emissive glow on a part
`explode`	Exploded view (whole car or single part)
`isolate`	Hide everything except one part
`reset_view`	Reassemble, restore visibility, reset camera
`show_specs`	Select part + info card

Named paint colors

Guards Red, Nardo Grey, Racing Green, Blueprint Blue, plus red, blue, green, black, white, silver, yellow, orange, purple, gold, pink, or any hex.

Example prompts

paint it guards red
make it Nardo grey
show the engine
explode the car
isolate the wheels
reset view
what's the horsepower? (uses Supabase / seed data)

Data tools (server only)

get_part_facts, get_overview, get_performance, get_pricing, search_knowledge, find_car

Supabase

Schema: supabase/schema.sql
Migration: supabase/migrations/20260626000000_blueprint_schema.sql
Tables: cars, car_variants, car_specs, performance_tests, pricing_market, research_chunks, etc.
RLS: public read, private write — anon/publishable key in clients, never service role in browser
Fallback: agent uses embedded seed data if Supabase is unavailable

Apply schema in Supabase SQL editor or via CLI. See supabase/README.md.

Duplicate Deno edge function lives under agent/supabase/functions/agent/ (same logic as api/agent.js for Supabase-hosted deploys).

Flutter (`integrations/blueprint-flutter/`)

Reference mobile client: car list, detail, compare, AI chat screen.

cd integrations/blueprint-flutter
flutter pub get
# Copy .env.example → .env with SUPABASE_URL + SUPABASE_ANON_KEY
flutter run

More detail: integrations/blueprint-flutter/README.md, SETUP_GUIDE.md, COMMANDS.md.

Environment variables

Copy .env.example → .env at repo root. Never commit .env.

Variable	Used for
`FEATHERLESS_API_KEY`	Featherless LLM
`LLM_PROVIDER`	`featherless` (default) or `openai`
`ATLAS_MODEL`	e.g. `Qwen/Qwen2.5-7B-Instruct`
`OPENAI_API_KEY`	If using OpenAI instead
`SUPABASE_URL`	Agent data lookups
`SUPABASE_SERVICE_ROLE_KEY`	Server-side agent only
`SUPABASE_ANON_KEY` / `SUPABASE_PUBLISHABLE_KEY`	Optional read client
`AGENT_PORT`	Local agent port (default `8000`)

Set the same vars in Vercel → Project → Environment Variables for production AI.

Vercel deployment

npx vercel deploy --prod

Config: vercel.json — static root ., serverless api/agent.js (10s max on Hobby)
Ignore large assets: .vercelignore excludes multi‑MB GLBs; they load from jsDelivr on *.vercel.app
CDN resolver: explorer/asset-config.js → resolveAsset('./path') rewrites to
https://cdn.jsdelivr.net/gh/ShadowEsu/Blueprint@main/explorer/...?v=gt3parts2
Showroom GLB (assets/blueprint-showroom-car.glb) is included in deploy (~33MB) for the launch hero

After pushing to main, jsDelivr picks up new explorer assets within a few minutes.

Project layout

index.html                 Launch page
src/main.js                Hero Three.js + scroll sections
src/styles.css
assets/                    Logo, Featherless SVG, showroom GLB

explorer/
  Garage.dc.html           Car picker
  Car Anatomy Explorer.dc.html   Main 3D viewer + AI
  Wind Tunnel.html
  asset-config.js          CDN path helper
  agent-bridge.js          Browser → /api/agent
  support.js               DC runtime (React + templates)
  assets/                  Monolithic car GLBs
  assets/gt3rs/parts/      GT3 RS multipart (114 GLBs)
  uploads/engine_teardown/ GT3 engine teardown

api/agent.js               Vercel serverless AI
scripts/serve-agent.mjs    Local AI dev server

supabase/                  Schema + migrations
integrations/blueprint-flutter/   Flutter app

npm scripts

Script	Command	Port
`npm run dev`	`python3 -m http.server 4173`	4173
`npm run dev:agent`	`node scripts/serve-agent.mjs`	8000
`npm run build`	No-op (static site)	—

Troubleshooting

Blank page / modules fail
Use npm run dev or the Vercel URL. Don't open HTML files directly (file://).

GT3 RS not showing on Vercel
Wait for 114 parts to load from jsDelivr (first visit is slow). Hard refresh. For instant loads, use local dev. Ensure GitHub main has the latest explorer/assets/gt3rs/parts/.

Assistant says unavailable
Run npm run dev:agent locally. On Vercel, check env vars and /api/agent returns 200.

Paint / colors don't change
Use the anatomy explorer orb (not wind tunnel). Try paint it red. Needs agent-bridge.js + scene handlers in Car Anatomy Explorer.dc.html.

Hero car missing on launch page
Hard refresh. Mobile shows a fast LaFerrari first, then upgrades to the full showroom model.

AI timeout on Vercel Hobby
Agent is tuned for ~8s responses. Very long prompts may hit the 10s function limit.

License / assets

3D models carry their own attributions in-app (Sketchfab CC BY 4.0 where noted). GT3 multipart assembly is Blueprint explorer assets.

Navigation Menu

Repository files navigation

BluePrint AI

What’s in this repo

Architecture (what we actually run)

System overview

User journey (pages)

AI assistant request flow

GT3 RS multipart load

Asset deployment split

Supabase data layer

Tech stack map

Quick start (local)

Live routes

Launch site (`/`)

Explorer (`/explorer/`)

Garage

Car Anatomy Explorer

GT3 RS model

Wind Tunnel

Engine teardown

AI assistant

How it works

Scene tools (client executes these)

Named paint colors

Example prompts

Data tools (server only)

Supabase

Flutter (`integrations/blueprint-flutter/`)

Environment variables

Vercel deployment

Project layout

npm scripts

Troubleshooting

License / assets

About

Releases

Packages

Contributors

Languages

Name		Name	Last commit message	Last commit date
Latest commit History 14 Commits
.cursor		.cursor
.vscode		.vscode
agent		agent
api		api
assets		assets
explorer		explorer
integrations/blueprint-flutter		integrations/blueprint-flutter
scripts		scripts
src		src
supabase		supabase
vendor		vendor
.env.example		.env.example
.gitignore		.gitignore
.vercelignore		.vercelignore
README.md		README.md
app.html		app.html
index.html		index.html
package-lock.json		package-lock.json
package.json		package.json
server.py		server.py
vercel.json		vercel.json

Folders and files

Latest commit

History

Repository files navigation

BluePrint AI

What’s in this repo

Architecture (what we actually run)

System overview

User journey (pages)

AI assistant request flow

GT3 RS multipart load

Asset deployment split

Supabase data layer

Tech stack map

Quick start (local)

Live routes

Launch site (/)

Explorer (/explorer/)

Garage

Car Anatomy Explorer

GT3 RS model

Wind Tunnel

Engine teardown

AI assistant

How it works

Scene tools (client executes these)

Named paint colors

Example prompts

Data tools (server only)

Supabase

Flutter (integrations/blueprint-flutter/)

Environment variables

Vercel deployment

Project layout

npm scripts

Troubleshooting

License / assets

About

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages

Launch site (`/`)

Explorer (`/explorer/`)

Flutter (`integrations/blueprint-flutter/`)