What we build
Most AI agents today are built for developers. Oro enables non-technical businesses to deploy autonomous AI employees without coding. Our mission is to bring AI employees to millions of SMBs across Asia-Pacific and India.
Oro Shin develops Oro AI Agent: a platform that lets any business owner, shop, restaurant, real-estate office, jewelry store, build their own AI Agent Employee Team in minutes, with no code. A Sales Agent that qualifies leads and closes conversations. A Support Agent that handles FAQs and payment verification 24/7. A Marketing Agent that generates and schedules posts automatically. An Operations Agent that updates your catalog, verifies slips, and sends weekly reports. All operating autonomously, across Facebook Messenger, Instagram, WhatsApp, Telegram, LINE and TikTok.
We focus on languages and markets the global AI tooling industry under-serves: Thai, Vietnamese, Indonesian, Filipino, Korean, Japanese, Traditional and Simplified Chinese, and Hindi & Tamil. Our AI agents understand local commerce conventions, lakh- and won-based pricing, regional holiday schedules, UPI, LINE Pay and local payment-slip flows, market-specific delivery norms, and reply to customers in their own language and tone.
Building our own AI
Oro Shin is an applied-AI company, not just an app. Beyond Oro AI Agent, we develop our own AI models in-house, including foundation models and research into AI that helps developers find and fix security vulnerabilities in their own code before attackers can exploit them.
Owning the full stack, agents, models, training data, and infrastructure, lets us deliver capabilities and economics off-the-shelf tools can't match, and keeps the AI serving our Asia-Pacific customers under our own control rather than a third party's. This investment in core AI research is what lets a small, focused team compete with far larger companies.
Founder
Than Kywal Nyein
Founder & CEO
Founded Oro Shin in 2024 with a single goal, bring enterprise-grade AI automation to small business owners in emerging markets, in their own language, on the channels their customers already use. Hands-on with the codebase, model selection, customer support, and business development.
Where we operate
Oro Shin operates from Bangkok, Thailand, with India go-to-market launching in 2026 (Bengaluru-based partner pilot in onboarding). We serve customers across Asia-Pacific, Thailand, Indonesia, the Philippines, Vietnam, Korea, Japan, Hong Kong, and Taiwan, and are expanding into India to bring the same multi-channel, multi-language agent to South Asia's SMBs.
Our production platform runs on Google Cloud, Cloud Run (auto-scales to zero when idle, up to hundreds of instances under load), Cloud SQL (PostgreSQL + pgvector) and Memorystore Redis, fronted by a Cloudflare global edge network, with E2B Firecracker microVMs for heavy on-demand compute and Alibaba Cloud model serving for our most economical tier. Per-tenant data isolation by default. We are backed by the Google for Startups Cloud Program, Alibaba Cloud Catalyst, AWS Activate, and E2B for Startups credit programs, a deliberately multi-cloud posture so no single provider outage or price change can take a customer's agent offline.
Partners & programs
Primary cloud
Google for Startups Cloud Program
Cloud Run (auto-scaling), Cloud SQL + pgvector, Memorystore Redis, Artifact Registry, Cloud Scheduler.
Model serving
Alibaba Cloud Catalyst
Model Studio (DashScope) serves our most economical Oro tier, fast, low-cost replies for everyday customer messages.
Cloud credits
AWS Activate
Multi-cloud redundancy and the India-region expansion path. Keeps us provider-independent at the infrastructure layer.
Sandbox compute
E2B for Startups, $20K + Pro Tier
On-demand Firecracker microVMs for heavy agent work, web scraping, document generation, data extraction, SEO crawls.
Program member (verified)
NVIDIA Inception
Verified member. GPU-backed inference planned for our Thai & Southeast / East Asian language fine-tune roadmap.
Status & traction
- Private beta: 50–100 paying B2B customers across food, retail, services, and content creation. Real Messenger + Instagram traffic in production.
- iOS: public TestFlight beta, join here
- Android: closed testing on Google Play, join by emailing orotest@googlegroups.com
- Oro Code shipped (May 2026): in-sandbox autonomous coding agent inside an E2B Firecracker microVM. The main AI delegates whole multi-step coding tasks, build a site, scaffold a backend, port a script, and Oro Code runs them end-to-end.
- Pure-edge architecture live (May 2026): inbound messages from every channel are handled at the Cloudflare edge in ~50ms, with Google Cloud Run assembling the full business context behind it. Production database migrated to Google Cloud SQL (PostgreSQL 16 + pgvector), TLS-required, row-level tenant isolation.
- Funding: bootstrapped, founder-funded
- Public launch: planned Q3 2026, Play Store + App Store + Microsoft Store
Your AI Agent Employee Team
Most “AI for business” tools are a single auto-reply script bolted onto one inbox. Oro AI Agent is a full AI Agent Employee Team that a non-developer can stand up in minutes. No code, no servers, no DevOps. You describe your business in plain language; each agent learns your products, voice, and rules, then operates autonomously, 24/7, across every social channel and your own website.
Together, these agents replace tasks that normally need a whole support, sales, marketing, and operations team: qualifying leads, following up, scheduling posts, verifying payment slips, updating catalogs, running SEO crawls, drafting emails, generating documents, and reporting back. All without you opening the app.
Sales Agent
Qualifies leads, answers product questions, quotes prices, follows up quiet prospects automatically, and closes conversations in the customer’s own language, 24/7, across every channel.
Support Agent
Handles FAQs, order status, payment-slip verification and escalation routing 24/7 across Messenger, Instagram, WhatsApp, Telegram, LINE and TikTok, with full conversation memory.
Marketing Agent
Generates and schedules posts, replies to all comments, recrawls your site for SEO keywords, and sends reach reports, automatically, on your posting calendar, in the right language for each market.
Operations Agent
Verifies payment slips, updates the product catalog, exports reports to Google Sheets or Excel, and runs the weekly business summary, without you opening the app.
Non-developers build in minutes
A shop owner creates an AI employee the way you'd send a chat message. An agent is just configuration: persona, channels, knowledge, skills, schedules, not code you have to deploy. Onboarding a new business changes zero infrastructure.
Autonomous, 24/7
Agents run on schedules and react to events on their own, auto-posting at 9am, recrawling your site for SEO, sending the Monday report, following up a quiet lead. No app needed.
Everywhere your customers are
Messenger, Instagram, WhatsApp, Telegram, LINE, TikTok and your Google Business Profile — one agent team, one shared memory, one brand voice. Customers can switch channels mid-conversation and the agents never lose the thread.
An auto-scaling platform, not a single server
Oro runs the way modern AI products run. All agents share one multi-tenant runtime that scales to zero when idle and to thousands of concurrent conversations under load. You pay for what an agent actually does, not for servers sitting idle. Four runtime tiers route every task to the cheapest place that can run it:
Tier 1 · Edge
Cloudflare Worker
Every inbound message and scheduled trigger lands on Cloudflare's global edge, near the customer, anywhere in the world. Per-request isolation, effectively unlimited concurrency.
Tier 2 · App
Google Cloud Run
In-app streaming chat and the heavy context assembler, loading an agent's persona, catalog, rules, and conversation history. Auto-scales 0→N; you pay nothing when no one is talking.
Tier 3 · Heavy compute
E2B Firecracker microVMs
Anything that needs a real computer, Python, a headless browser, PDF/Excel parsing, SEO crawls, document generation, spawns a sandbox on demand, runs the task, and is destroyed. Billed per second of actual compute.
Tier 4 · Long-running
Cloud Run Jobs & Scheduler
Super-heavy or long batch work, large rebuilds, and every cron-driven autonomous task. One scheduler entry per agent task fires the edge runtime on time, every time.
Data & model layer
Every agent's configuration, catalog, conversations, and long-term memory live in Cloud SQL (PostgreSQL + pgvector) with row-level tenant isolation; Memorystore Redis handles rate-limiting, idempotency, and live chat deltas; uploads sit in Cloudflare R2. Behind a single router, the fastest tier runs on Alibaba Cloud model serving and the premium reasoning tiers route through OpenRouter, with Voyage AI for embeddings, swappable per agent with zero code change, so a provider outage or price spike never strands a customer.
How it works under the hood
Oro AI Agent isn't a thin wrapper around one LLM API. It's a multi-model orchestrator with production-grade reliability features that small businesses normally only get from enterprise contracts.
Multi-model routing
Every customer message routes through one of 8+ AI models (Gemini Flash, Claude, OpenAI, DeepSeek V4, Qwen 3.6, Kimi K2.6, Gemma 4) picked per task on cost and quality. Cheap routing for “what's your address?”, premium reasoning for negotiation. Owners pay up to 70% less than running everything on a single top-tier model.
Auto-scaling compute (Oro Advance)
Heavy AI tasks (Auto Learn product extraction, deep analytics, custom long-running workflows) run in isolated Firecracker microVMs (powered by E2B) that scale from zero to dozens of instances on demand. Owners pay per turn, not per uptime. A page can spike to 50 concurrent agents and back down in seconds with no operator action.
Continuous conversation state
The agent maintains a slot-filled memory of every customer (phone, address, item, quantity, payment status, order context) across message bursts and sessions. The AI never re-asks for collected info or loses chain of thinking after 3-4 turns. Works across Messenger, Instagram, and Telegram on the same customer identity.
Reply Auditor
Every AI reply runs through a second-pass verifier that catches missed questions, factual drift, and tone mismatches before the customer sees them. If the audit fails, the system auto-sends a polished follow-up in the customer's language. Most businesses never know it ran, they just see fewer escalations.
Massive parallelism
One agent handles thousands of customer conversations in parallel with no per-conversation cost overhead. A page with 10,000 daily inquiries pays the same per-message rate as a page with 10. The webhook pipeline auto-replies in roughly 1.5 seconds on average, faster than most human admins typing.
Prompt cache & token economics
The system reuses cached prompt prefixes across the page's entire catalogue, rules and persona, so a repeat query inside the cache window is billed at a fraction of the input cost. Combined with multi-model routing, that is what lets the cheapest tier stay cheap.
Multi-tenant isolation
Every business gets its own database schema, storage, and embedding index. Your customer list, product catalog, and conversation history never share runtime memory with another business's queries. Per-tenant encryption keys; AsyncLocalStorage-scoped tenant context end-to-end.
Auto Learn
On first connect, the agent reads up to 100 of your recent posts, top reviews, page about, and 20 conversations. It extracts products with prices in any currency (lakh, baht, rupee, won, yen, USD), rules, FAQs, and your brand voice. ~40 seconds for a 100-post page. Re-runs cleanly with no duplicates.
Built-in tool calling
Admins talk to Oro Chat to manage everything: over 100 tools to create agents, message customers across Messenger, Instagram, WhatsApp, Telegram, LINE and TikTok, schedule posts, manage products and orders, publish a website and generate reports. No menus to memorise.
Bring your own key
Higher-volume pages can plug in their own Anthropic, OpenAI, or Gemini API key. Pay the AI provider directly, keep our platform features. Capacity becomes effectively unlimited; we charge nothing for tokens you bring yourself.
Oro Code, autonomous engineer
When you ask for something multi-step (“build me a 3-page site, deploy it, verify it loads”), the main AI delegates the whole task to Oro Code, an autonomous coding agent running inside its own Firecracker microVM. It writes files, runs commands, deploys to the cloud, and returns one summary. No copy-paste, no DevOps, no console.
One-click cloud deployment
Describe what you want and Oro ships it: a permanent website to <slug>.oroaiagent.com on the global edge, a webhook / API endpoint, or a full-stack app on auto-scaling Cloud Run. The user never sees a console, a CLI, or a YAML file.
Five channels, one agent
Messenger, Instagram (DM + comment auto-reply + scheduled posts), Telegram, LINE, and WhatsApp Business all reach the same agent with the same conversation state and persona. Customers can switch channels mid-thread; the AI doesn't lose context.
Live sandbox view
When Oro is working on a long task, you watch it happen: terminal output, file tree, live browser preview, deployment pipelines, payment-slip vision results. Built so the owner can trust what the agent is doing, not merely what it claims it did.
Zero-key sandbox security
No provider API keys ever enter a user's sandbox. Each microVM carries only a 15-minute scoped JWT and reaches the AI model through our authenticated proxy. Rate-limited, spend-capped, audit-logged. Compromise of a sandbox leaks nothing useful.
On the integration side, the platform connects natively to Facebook Messenger, Instagram, WhatsApp Business, Telegram, LINE, TikTok, YouTube and Google Business Profile, and to Google Drive, Docs and Sheets. Catalogues can be imported from Excel, Word, CSV or PDF in any language: the agent extracts products, prices and stock automatically, and exports back to any of them.
Contact
General support
support@oroaiagent.comAndroid testers
orotest@googlegroups.comWebsite
oroaiagent.comOro Shin is an independent third-party tool. Not affiliated with Meta Platforms, Inc., Google LLC, Apple Inc., or any other platform we integrate with. © 2026 Oro Shin. All rights reserved.