An operations director in Kwun Tong watched OpenAI's DevDay keynote on 29 September and wanted exactly what was on stage: an AI agent that keeps working overnight, watches the inbox, and has a draft ready by 9am. Then she tried to sign up from a +852 number.
OpenAI Dots are not available to Hong Kong companies directly. That leaves a real buying decision: wait, route through an overseas entity, or choose an always-on agent that does work here. This page compares the options on price, control and fit, including where UD's own products are the wrong choice.
Can Hong Kong companies use OpenAI Dots?
Not directly. OpenAI has not served Hong Kong since July 2024, and accounts cannot be registered with a +852 number or Hong Kong IP address. Dots also require ChatGPT Pro, Business Premium or an admin-enabled Enterprise beta. The only compliant route is a group contract held by an entity in a supported region.
According to eesel’s breakdown of OpenAI’s launch documentation, Dots launched on 29 September 2026. Each dot runs on GPT-6 Astra with its own cloud computer and browser, connects to more than 4,000 apps, and keeps working between conversations.
Access is narrow even where OpenAI operates. The first dot is included on ChatGPT Pro, from US$100 a month, and on Business Premium seats. Enterprise workspaces get a beta that stays off until an admin enables it. PTS Consulting’s 2026 review confirms Hong Kong remains outside OpenAI's supported list.
If your company is part of a multinational with an OpenAI Enterprise contract in Singapore, the UK or the US, ask group IT whether Hong Kong staff are covered. Do not rely on VPN workarounds: they breach OpenAI's terms and create an audit problem.
What should an always-on AI agent replacement actually do?
A credible Dots alternative should run continuously without an open chat window, keep durable memory across sessions, connect to the messaging and productivity tools your team already uses, ask for approval before sensitive actions, and let you control where data and model calls go. Price matters less than control in enterprise pilots.
Evaluation criteria used on this page--- Persistence: does it keep working when nobody is chatting?
--- Memory governance: can you inspect, edit or delete what it remembers?
--- Connectors: WhatsApp, Slack, Teams, Gmail, Workspace or Microsoft 365?
--- Approval controls: does it pause before sending, paying or deleting?
--- Data location: where does the agent run, and which model provider sees prompts?
--- Availability in Hong Kong: can a Hong Kong entity buy and pay for it legitimately?
One detail from OpenAI's own documentation sets the bar: Dots users currently cannot view, edit or delete individual dot memories. Any alternative that lets you inspect memory is already ahead on governance.
What are the best OpenAI Dots alternatives in Hong Kong?
The realistic shortlist for Hong Kong organisations is Google Gemini Spark, self-hosted Hermes Agent or OpenClaw, and managed versions of those open-source agents such as UD's Hermes Agent Express and OpenClaw Express. Multinationals may also access Dots through an overseas group contract. Each fits a different organisation profile.
1. Google Gemini Spark. Google launched Spark in Hong Kong on 29 July 2026, according to The Standard. It runs in Google's cloud, connects natively to Gmail, Docs, Sheets and Calendar, keeps working when your laptop is closed, and asks for consent before high-risk actions such as transactions or sending email. It launched for Google AI Ultra subscribers, with AI Pro access rolling out afterwards. 2. Self-hosted Hermes Agent. Hermes Agent is Nous Research's open-source agent, released under the MIT licence in February 2026. It keeps persistent memory and self-written skills in local storage on your own server, connects to Telegram, Slack, Discord and other channels, and works with whichever model you point it at. 3. Self-hosted OpenClaw. An open-source, self-hosted assistant that acts through chat platforms such as WhatsApp, Telegram and Slack. Like Hermes, the software is free; your team owns the server, updates and security. 4. Hermes Agent Express (UD). A managed deployment of Hermes Agent on Hong Kong cloud servers, provisioned in about three minutes according to the product page, with 15+ messaging platforms, 20+ model providers and Cantonese technical support. 5. OpenClaw Express (UD). A managed OpenClaw deployment on isolated Hong Kong cloud servers with models pre-loaded, as described on the product page. 6. Dots via an overseas group contract. Only viable for multinationals whose OpenAI agreement already covers Hong Kong users.How much do OpenAI Dots alternatives cost in Hong Kong?
At the time of writing, Gemini Spark launched on Google AI Ultra at HK$1,588 per month. UD's Hermes Agent Express and OpenClaw Express are listed from HK$220 per month for Starter and HK$580 per month for Pro, against list prices of HK$488 and HK$1,288. Self-hosted options cost server time plus model usage.
Published price facts (checked 6 October 2026)--- OpenAI Dots: first dot included with ChatGPT Pro from US$100/month or Business Premium; not sold to Hong Kong entities (source).
--- Gemini Spark: Google AI Ultra, HK$1,588/month at Hong Kong launch; AI Pro HK$158/month, with Spark access rolling out to Pro after launch (DCFever, Hong Kong pricing).
--- Hermes Agent Express Starter: listed from HK$220/month (struck-through HK$488). 2 vCPU, 4GB RAM, 80GB SSD, single-user agent, 2 connected platforms.
--- Hermes Agent Express Pro: listed from HK$580/month (struck-through HK$1,288). 6 vCPU, 16GB RAM, 240GB SSD, unlimited platform connectors.
--- OpenClaw Express: same price points, HK$220 and HK$580 per month, on 2 vCPU/4GB and 6 vCPU/16GB servers.
--- Promo code: both UD pages display code new2026 for 50% off. The pages do not state whether it stacks with the listed price; confirm at checkout.
--- Not included: model usage. Managed agents connect to providers such as OpenRouter, Anthropic or OpenAI APIs, which bill separately per token.
For a worked total-cost example, see our Hermes Agent Express pricing breakdown and OpenClaw hosting cost analysis.
Which Dots alternative is right for which type of organisation?
Choose Gemini Spark if your company already runs on Google Workspace and wants the least setup. Choose a managed open-source agent if you need WhatsApp or Slack reach, model choice and a predictable monthly server fee. Self-host only if you have engineers to maintain it. Multinationals should first check existing group OpenAI contracts.
Verdict by buyer type--- Google Workspace organisation, 50 to 300 staff: Gemini Spark. Native Gmail and Calendar access outweighs model choice.
--- Microsoft 365 organisation with heavy WhatsApp customer traffic: a managed Hermes Agent or OpenClaw deployment, because Spark's strength is Google's own apps.
--- Team that wants model independence: Hermes Agent, managed or self-hosted, since it works with 20+ providers and avoids single-vendor lock-in.
--- Organisation with an in-house platform team: self-host. Software is free and data stays on your servers, but budget engineering time for updates and security.
--- Multinational with an overseas OpenAI agreement: confirm coverage for Hong Kong users before buying anything new.
--- Regulated workload handling client personal data: none of these should go live before a PDPO assessment and a decision on which model provider receives prompts.
What are the limitations of UD's Hermes Agent Express and OpenClaw Express?
UD's managed agents solve deployment, not governance. The published plans do not list single sign-on, role-based access or audit exports; model usage is billed separately; prompts sent to overseas model APIs leave Hong Kong; and the Starter tier supports only two connected platforms. Gemini Spark wins on zero-setup Workspace integration.
Where UD is the wrong buy--- You need enterprise identity controls today. SSO, role-based permissions and centralised audit logs are not listed on either product page. Ask before assuming.
--- You live in Google Workspace. Spark connects to Gmail and Docs without configuration. A managed open-source agent needs connectors set up.
--- You want one fixed price including AI usage. Server fees are fixed; token costs vary with workload and model.
--- You need strict data residency. The server is in Hong Kong, but if the agent calls an overseas model API, prompt content leaves Hong Kong. Only a locally hosted model avoids that.
--- You plan to run WhatsApp at scale. Confirm whether the connection uses Meta's official Business API; unofficial bridges carry account-ban risk.
These limits are not unusual for open-source agents. They are the questions that separate a productive pilot from a stalled one.
How should you pilot an always-on AI agent safely?
Run a 30-day pilot on one workflow with read-only access first, require approval for every outbound message, choose a model provider your compliance team has cleared, and measure hours saved against a baseline. Expand write permissions only after two clean weeks. This approach works for Dots, Spark and open-source agents alike.
--- Week 1: read-only access to one inbox or channel; the agent drafts, humans send.
--- Week 2: add scheduled tasks, such as a daily 8am briefing of overnight enquiries.
--- Weeks 3 and 4: allow pre-approved actions for low-risk replies; keep payments, deletions and external emails on manual approval.
--- Measure: hours saved per week, error rate on drafts, and staff adoption.
A 20-person customer operations team saving even 30 minutes per person per day recovers roughly 50 hours a week, which is the number your CFO will ask for.
What is the right next step for a Hong Kong team that wanted Dots?
Decide which ecosystem you live in, then pilot. Google Workspace teams should trial Gemini Spark. Teams needing WhatsApp, Slack and model choice should pilot a managed Hermes Agent or OpenClaw deployment on one workflow. Everyone should settle model-provider and PDPO questions before granting an agent write access.
Always-on agents will not stay a novelty. The organisations that learn how to govern them this quarter will scale them next year, while others are still waiting for a product that may never launch here.
You do not need to wait for OpenAI to arrive. We understand the cold edges of AI and the hard parts of your work, and UD has walked with Hong Kong enterprises for twenty-eight years, making technology a partnership with warmth.
Reviewed by the UD enterprise AI team. Prices checked against published product pages on 6 October 2026.Put an Always-On Agent to Work in Hong Kong
Ready to pilot an always-on AI agent without waiting for Dots? We'll walk you through every step, from choosing the plan and model provider to connecting your channels and setting approval rules. Your agent can be live on Hong Kong servers in minutes.
