How Much Does Hermes Agent Express Cost?
Hermes Agent Express is sold in two monthly tiers on its product page: a Starter plan from HK$220 per month against a list price of HK$488, and a Pro plan from HK$580 per month against a list price of HK$1,288.
That is the whole published price list. There is no third tier, no per-seat charge and no setup fee shown.
Plans and limits, as published:
--- Starter, from HK$220/month (list HK$488). 2 vCPU, 4GB RAM, 80GB SSD. Persistent memory. 2 connected platforms. Positioned for solo creators, freelancers and self-employed professionals.
--- Pro, from HK$580/month (list HK$1,288). 6 vCPU, 16GB RAM, 240GB SSD. Agent-curated memory at scale, unlimited platform connectors, enterprise data isolation. Positioned for teams, SMEs and cross-departmental workflows.
--- Included on both: deployment in about 3 minutes, connection to 15+ messaging platforms including Telegram, Slack, WhatsApp, Discord, Signal and email, access to 20+ LLM providers, Hong Kong cloud servers, local payment methods, and technical support in Cantonese and Traditional Chinese.
Annual arithmetic at list-facing prices: Starter is HK$2,640 a year, Pro is HK$6,960 a year. Against the struck-through prices those same plans would be HK$5,856 and HK$15,456.
Is the 50% Discount Code Real, and What Does It Apply To?
Yes, the offer is published on the page itself. The banner reads: enter code new2026 to get 50% off. It sits directly above both plan cards, in both the English and Traditional Chinese versions of the page.
What the page states, and what it does not:
--- Amount: 50% off, stated plainly.
--- Redemption: enter the code new2026. No account or eligibility step is described beyond that.
--- Eligibility: not stated. The page does not say new customers only, nor limit it to a particular plan.
--- Expiry: not stated. The banner says "limited offer" but carries no end date.
--- Exclusions: not stated.
--- Renewal price: not stated. The page does not say whether the discount applies to the first month, the first year, or the life of the subscription.
Both prices are also written as "from", which normally signals a starting figure rather than a fixed one.
The practical consequence is one question to ask before paying. Ask whether HK$220 is already the discounted figure or whether the code takes it lower, and ask what the price becomes at renewal. If the code halves HK$220 again, the first-year cost is HK$1,320 instead of HK$2,640. That is a HK$1,320 difference decided by a sentence nobody has written down yet, so get the answer in writing.
What Is Not Included in the Monthly Price?
The subscription buys the server, the deployment and the platform gateway. It does not buy the intelligence. The page describes 20+ LLM providers under the heading "bring your own model", which means the AI usage arrives on a separate invoice from a separate company.
This is the single most common surprise in agent hosting, and it is not unique to this product. Every self-hosted or managed agent works this way unless the vendor explicitly bundles model usage, and this page does not state that it does.
Why it matters more for an agent than for a chatbot. A chatbot answers when spoken to. An always-on agent with persistent memory re-reads its own history, runs scheduled tasks and works in the background, so it can consume model capacity on days nobody talks to it at all.
A concrete way to see the gap. Two businesses can pay the identical HK$220 subscription and receive model invoices that differ by an order of magnitude, because one uses the agent for a handful of daily messages and the other has it summarising every inbox thread on a schedule. The subscription is fixed. The part that varies is the part nobody quotes you.
Before committing, the useful exercise is to estimate your own model spend rather than accept a headline. UD's own comparison of real cost per task across GPT-5.6, Gemini 3.7 Flash and Claude Sonnet 5 is a more honest starting point than any vendor estimate, including this one.
Two other costs are worth naming because no hosting plan removes them. Messaging platform fees still apply where the platform charges, most obviously WhatsApp. And your own time in the first fortnight, connecting accounts and correcting the agent, is real even when deployment takes three minutes.
Starter or Pro: Which Tier Actually Fits?
The deciding number is not RAM. It is the platform connector count, because that is the only hard functional limit between the two tiers.
Starter caps you at 2 connected platforms. If your agent lives on WhatsApp and Telegram, that is enough. If you want WhatsApp, Telegram, Slack and email in one gateway, Starter cannot do it at any traffic level, and the upgrade is a functional necessity rather than a performance one.
Pro is where team use begins. Unlimited connectors plus enterprise data isolation is the combination that matters once more than one person's conversations are in the same memory store. A 6-person firm putting client conversations into a shared agent memory should treat data isolation as the reason to buy Pro, not the extra vCPUs.
The spec jump between the tiers is threefold on CPU, fourfold on RAM and threefold on storage, for 2.6 times the price. That ratio is reasonable, but it only pays for itself if you are actually running long memory or several connectors.
A simple test: count the messaging channels your business genuinely uses daily. Two or fewer and one user, start with Starter. Three or more channels, or more than one person, and you are buying Pro whether you like it or not.
Where Hermes Agent Express Is the Wrong Buy
An honest comparison has to name the cases where a reader should close the tab. There are four here, and one of them is the underlying software itself.
You have IT capability and a spare server. Hermes Agent is an open-source project from Nous Research, released in February 2026 under an MIT licence, and it is designed to be self-hosted (NousResearch/hermes-agent on GitHub). If someone in your company can provision a server, configure Python and wire up bot APIs, the software costs nothing and you pay only for hosting. What you are buying from UD is the removal of that work, not the agent.
You want a customer-facing chatbot. This is a personal or team agent that lives in your messaging apps. It is not a website chat widget with a knowledge base and a dashboard for enquiry volume. Those are different products with different pricing shapes.
You need a written all-in monthly figure before approval. Because the model spend is separate and the discount terms are unstated, no honest quote for this product is a single number today. If your approval process needs one number, get both the subscription terms and a capped model budget in writing first.
You are comparing on price alone. Because Hermes Agent is open source, other providers host it too. FlowEngine advertises managed Hermes Agent hosting from US$12 per month, roughly HK$94, which is below the Starter tier here. What UD adds against that is Hong Kong servers, local payment methods and support in Cantonese and Traditional Chinese. If none of those three matter to your business, the cheaper host is the rational purchase.
You want to compare UD's own options. UD sells a second managed agent product, OpenClaw Express, aimed at a different workflow. The enterprise-side breakdown of what OpenClaw costs a Hong Kong team is worth reading before deciding which of the two you actually want.
The Verdict by Buyer Type
Different businesses reach different answers from the same price list, so here is the recommendation split by who is asking.
--- Solo professional, one or two channels, no IT skills. Starter at HK$220 a month is the intended fit, and the discount code makes the first commitment small. Confirm the renewal price before subscribing.
--- Small team of 3 to 15 with shared client conversations. Pro, for data isolation and unlimited connectors. Budget the model spend separately and review it after month one.
--- Company with an internal IT person and existing servers. Self-host the open-source project first. If it takes more than a week of that person's time, the managed plan was cheaper all along.
--- Business whose real need is customer enquiries at scale. Buy nothing here. A knowledge-base chatbot with published conversation tiers is the correct shape of product for that job.
--- Anyone who has not yet decided what the agent is for. Buy nothing this month. An always-on agent with persistent memory rewards a clear job description and punishes a vague one, because it will keep remembering whatever you feed it.
Frequently Asked Questions
Is there a free trial?
The product page does not advertise one. It offers a discount code and a "Contact Our Experts" enquiry form, not a trial. If a trial matters to you, ask for it explicitly rather than assuming.
Is there a minimum contract length?
The page states no minimum term. That is a genuine advantage over subscription products that require a six-month commitment, but confirm it at checkout because an unstated term is not the same as a stated absence of one.
Where is my agent's data stored?
The page states Hong Kong cloud servers, and Pro adds enterprise data isolation. If your business handles client information under a confidentiality duty, ask for the storage location and retention period in writing before uploading anything.
Can it really reply in Cantonese?
Support is offered in Cantonese and Traditional Chinese, and the agent's language ability comes from whichever model you connect. Test your intended model on your own Cantonese phrasing during setup rather than trusting a general claim.
Does the price include the AI model?
No. The page describes 20+ LLM providers under a bring-your-own-model approach and does not state that any model usage is included, so budget for a second invoice from whichever provider you connect.
What happens to the agent's memory if I cancel?
Not stated on the page. Since the entire value of this product is accumulated memory, ask what happens to that memory on cancellation before you accumulate a year of it.
The Next Step
The published price is clear, the offer is real, and the gaps are all in the same place: what happens after the first invoice. Every question worth asking here is about month two, not month one.
Before you subscribe, get three answers in writing: whether HK$220 is before or after the code, what the price becomes at renewal, and what happens to your agent's memory if you stop. A vendor who answers all three quickly is one worth buying from.
We understand AI. UD stands with you. After twenty-eight years in Hong Kong, the questions we most want customers to ask are the ones about month two, because that is where every disappointing subscription actually goes wrong.
Reviewed by the UD AI team.
Talk to Us Before You Subscribe
If you want to know which tier fits your channels and what your model spend is likely to be, ask us before you pay rather than after. We will walk you through it step by step, from choosing the plan to connecting the first platform.
