Taksu Tech

Robots made in Bali by Taksu Tech, part of Ezra Robotics: the company profile, the reception robots for Balinese offices and resorts (their site, the screen app they run, and its speech service), and the education robots (their store and block simulator), with the Docker stack that runs them all on a Mac mini.

Repositories · Features · System · Run it · Deploy · Releases · Documentation · Develop

---

Taksu Tech sells two lines of robots, each with its own site because the markets differ: reception robots made for Bali (Snow and Timo in a Bali Edition, and Taksu Kiosk, a totem built in-house) and education robots sold across Indonesia. This repository is the deployment root: the infrastructure and the shared packages at the top level, every app in its own repository, pinned here as a submodule.

Repositories

Path	Repository
apps/web	company-profile	the company profile: services, works, blog, demos, contact
apps/robot	reception-robot-web	the reception robot site: line-up, robot pages, comparison, Aksara Bali, demos
apps/kiosk	reception-robot-kiosk	the screen app on Snow, Timo, and the Taksu Kiosk totem
apps/edu	education-robot-web	the education robots' store: robots, Drops, accessories, learning paths, business
apps/sim	education-robot-sim	the block simulator for Ciko and Momo, with Arduino C++ output
services/speech	reception-robot-speech	Whisper and Piper for the kiosk, on CPU

Features

Company

• The company profile in Indonesian and English: both product lines, custom robots, PCBs, IoT, and web apps, the works page, an MDX blog with RSS, demos, and a contact form that hands the message to WhatsApp or email.

Reception robots

• A site with the line-up drawn to the robots' real heights, a page and a demo per robot, a comparison, the Aksara Bali regulation with a live sign, and today's Pawukon day.
• One screen app for the 13.3-inch screens on Snow and Timo and the 24-inch portrait totem: queue tickets, a guest book, directions on the floor plan, the venue's organisation, Balinese customs, and Aksara Bali above the Latin on every sign.
• An assistant that answers from the venue's data, offline or through Ollama (native on the Mac mini), Anthropic, or any OpenAI-compatible API; Whisper and Piper for listening and speaking, with the robot's mouth following the voice; tickets and passes on 80 mm thermal paper.

Education robots

#	Robot	Starts at	For	Status
01	Ciko	age 7	a first robot car, ready to program	pre-order
02	Momo	age 11	a modular ESP32 robot that rebuilds four ways	pre-order
03	Garu	age 15	a ROS 2 rover with a camera, LiDAR, and an arm	coming soon
04	??		the next Drop from the R&D bench	revealed at launch

• A store with a page per robot, packages and add-ons, accessories, upgrade packs, arenas, and the Drops; a cart that turns the order into one WhatsApp message or email, with no payment on the site; bulk orders, training, and OEM for schools and teams.
• A free learning path per robot, each step one idea with a simulator mission or a guide, and a block simulator for Ciko and Momo with real sensors, 18 missions, and the Arduino C++ each program becomes.

Operate

• One Docker image per app, the speech service, and Caddy behind the existing Cloudflare Tunnel; Prometheus, Loki, Alloy, and Grafana with a ready dashboard; CI here and in every app's repository; images on GHCR.

System

flowchart LR
    U[Visitors, venues,<br/>robot screens] --> T[Cloudflare Tunnel]
    T --> C[Caddy<br/>127.0.0.1:4480]
    C --> W[web<br/>company profile]
    C --> R[robot<br/>reception robot site]
    C --> K[kiosk<br/>screen app]
    K --> S[speech<br/>Whisper, Piper]
    K --> O[Ollama<br/>native on macOS]
    C --> E[edu<br/>education robot store]
    C --> M[sim<br/>block simulator]
    E -- open the mission --> M
    W & R & K --> P[(packages<br/>content, ui, bali)]
    E & M --> Q[(packages<br/>catalog, edu-ui)]

Loading

Path
packages/content/	every fact the sites show: brand and names, both product lines, robots, venues, works, demos
packages/ui/	shared components and two themes: company.css (the company profile) and theme.css (the reception line and the kiosk, Bali)
packages/bali/	Aksara Bali transliteration, Balinese numerals, greetings, the Pawukon calendar
packages/catalog/	every fact the education robots' store and simulator show: robots, prices, accessories, Drops, learning paths, guides, missions
packages/edu-ui/	the education robots' theme (taksu.css), wordmark, race-number plate, components
packages/typescript-config/	shared TypeScript settings
docker/Dockerfile	one standalone Next.js image per app
docker-compose.yml, infra/caddy/	the stack: web, robot, kiosk, edu, sim, speech, and Caddy on the loopback
infra/monitoring/, infra/ollama/	Prometheus, Loki, Alloy, Grafana; the taksu-asisten model
infra/deploy.sh, infra/scripts/	deploy with a smoke test, and the cloudflared hostnames

Run it

git clone --recursive https://github.com/DarmawanWeb/TaksuTech.git && cd TaksuTech
pnpm install        # also installs the commit hooks, here and in each app and the speech service
pnpm dev            # web :3000, kiosk :3001, sim :3002, edu :3003, robot :3004
pnpm check          # Biome, typecheck, tests
pnpm test:speech    # the speech service's tests (Python, ffmpeg)

Without the speech service the kiosk listens and speaks through the browser; to try Whisper and Piper, run it on port 8000 and set SPEECH_URL=http://localhost:8000 in apps/kiosk/.env.local (how). Set NEXT_PUBLIC_WHATSAPP_NUMBER in apps/edu/.env.local to try the store's WhatsApp orders.

Deploy on the Mac mini

Everything runs in Docker; only Caddy opens a port, 127.0.0.1:4480. Ollama stays native on macOS to use the Metal GPU, and the containers reach it at host.docker.internal:11434.

git clone --recursive https://github.com/DarmawanWeb/TaksuTech.git ~/web/taksu-tech
cd ~/web/taksu-tech
cp infra/.env.example ~/web/taksu-tech.env     # hosts, CLOUDFLARE_TUNNEL, GRAFANA_ADMIN_PASSWORD
infra/ollama/setup.sh                           # once: the taksu-asisten model
ENV_FILE=~/web/taksu-tech.env infra/deploy.sh   # pinned apps, build, start, smoke test
ENV_FILE=~/web/taksu-tech.env CLOUDFLARED_CONFIG=~/.cloudflared/config.yml infra/scripts/cloudflare-ingress.sh

.github/workflows/deploy-macmini.yml deploys every green main from a self-hosted runner.

Hostnames and configuration

Hostname (default)	Goes to
taksu.agus-darmawan.com	web, the company profile
taksu-robot.agus-darmawan.com	robot, the reception robot site
taksu-kiosk.agus-darmawan.com	kiosk, with the microphone allowed
taksu-edu.agus-darmawan.com	edu, the education robots' store
taksu-sim.agus-darmawan.com	sim, the block simulator
taksu-grafana.agus-darmawan.com	Grafana; protect it with Cloudflare Access

Every hostname points cloudflared at the one target http://127.0.0.1:4480, and Caddy routes by host. The ingress script backs up config.yml, inserts the rules before http_status:404, validates, creates the CNAMEs, and restarts cloudflared.

Variable (infra/.env)	What it does
WEB_HOST, ROBOT_HOST, KIOSK_HOST, EDU_HOST, SIM_HOST, GRAFANA_HOST	the hostnames above
CADDY_PORT	Caddy on the loopback, 4480; never 80 or 8080 on this Mac mini
MONITORING	1 also runs Prometheus, Loki, Alloy, blackbox, and Grafana
CLOUDFLARE_TUNNEL	the existing tunnel, for the ingress script
WHATSAPP_NUMBER	where enquiries and store orders go; empty means email only
CONTACT_EMAIL	the store's email orders; defaults to brand.contact.email in packages/catalog
TOKOPEDIA_URL, SHOPEE_URL	the store's marketplace pages; each button hides while empty
AI_*, ASSISTANT_RATE_LIMIT	the kiosk's assistant: provider, model, timeouts, concurrency, rate limit
WHISPER_*, PIPER_VOICE_*	the speech service's model and voices
GRAFANA_ADMIN_USER, GRAFANA_ADMIN_PASSWORD	Grafana's own login

Public values are inlined into the images at build time, so change them and deploy again. On the first start the speech service downloads Whisper small and the Piper voices into the speech_models volume; until then the kiosk uses the browser's voice. With MONITORING=1, Prometheus scrapes Caddy and probes web, robot, kiosk, edu, sim, speech, and Ollama; Alloy ships every container's log in the taksu project to Loki; Grafana opens on the Taksu Tech: layanan dashboard.

Releases

Version
v2.0.0	Taksu Tech in repositories, every app pinned here side by side; the education line rebuilt as named robots (Ciko, Momo, Garu) with a store, Drops, learning paths, and a simulator, built under the working name ULIK; the company profile tells the new line.
v1.0.0	The MVP as one monorepo: the company profile and blog, Taksu Edu with kits per school level and a curriculum, the reception robot site and the kiosk with a voice assistant, the simulator, Aksara Bali, Docker and monitoring.

The first 274 commits are that monorepo, unchanged; every app's repository carries its own part of that history. refactor: split the company profile, robot site, kiosk, and speech service into repositories starts v2, and the education robots' rebuild joins it through the merge refactor: bring the store and the simulator into apps, next to the other apps.

Documentation

For	Document
Contributors	CONTRIBUTING.md: working with the submodules, house rules, blog posts, works
Company profile	company-profile/docs: architecture, development
Reception robot site	reception-robot-web/docs: architecture, development
Kiosk	reception-robot-kiosk/docs: screens, assistant, voice, printing, venues
Speech service	reception-robot-speech/docs: engines, API, voices
Education store	education-robot-web/docs: architecture, development
Simulator	education-robot-sim/docs: architecture, development

Develop

git submodule update --init --recursive
pnpm install        # husky here, and the same hooks inside each app and the speech service

Change an app inside its submodule and push it there, then commit the new pointer here (chore: update kiosk); Dependabot opens that pull request weekly too. The ci workflow checks out the submodules and runs Biome, the typecheck, the tests, the builds, and the speech tests, and validates the compose file, the Caddyfile, and the shell scripts; each app's repository has its own CI. docker.yml publishes ghcr.io/darmawanweb/company-profile, reception-robot-web, reception-robot-kiosk, education-robot-web, and education-robot-sim on main and v* tags.

Brand

• Names: packages/content/src/names.ts. The reception line's own name is still to be decided, so every site reads it from there; public copy carries the Taksu Tech label only, with no OEM names.
• Company profile: company.css, plain on purpose: white and petrol #12343b, cyan #1fb0cf for small accents, Hubot Sans and Mona Sans. The mark is a T drawn as a PCB trace that turns at 45° and ends in a via.
• Reception line: theme.css, made for Bali: Paras stone, Endek indigo, Malam night, Prada gold, Subak water; Aksara Bali above the Latin, as Pergub Bali 80/2018 asks. It is generated, so official signage is reviewed by a Balinese language adviser.
• Education robots: taksu.css in packages/edu-ui, Ciko's livery: red #e3262d for actions, asphalt #1f2228, chalk #f4f5f7, eye cyan #44ddeb for learning, signal amber #ffb020 for Drops; Anybody for headlines and race numbers, Atkinson Hyperlegible Next for text. python3 packages/edu-ui/scripts/wordmark.py <name> redraws the wordmark.
• Claims: no TKDN percentage for Taksu Kiosk before certification. piper-tts is GPL-3.0 and runs as its own service; check each voice's licence before commercial use.

Proprietary. Copyright (c) 2026 Taksu Tech, part of Ezra Robotics. All rights reserved.