A general chatbot tab is the fastest AI surface. A multi-tenant assistant is the fastest agent-shaped product. A private agent in your messenger — your own server, consented context, routines, and approval before anything sensitive is sent — is what Ermes makes easy. Pick the one whose tradeoffs match your week.
Short version. If you want answers in a tab, open a chatbot. If you want quick agent-shaped workflows on shared infrastructure, sign up for a multi-tenant assistant. If you want an AI agent that is yours — one server, one memory store, one human, and approval before anything sensitive is sent — start with Own your agent to see how the ownership, safety and easy-setup pieces fit together.
ChatGPT, Claude, Gemini and the rest take about a minute. You get a question-and-answer surface that lives in a browser tab and forgets between sessions. It is the fastest path to an answer. It is not an agent — it will not act on your tools, will not keep a routine, and will not text you on Friday afternoon.
Fit for: One-off questions, drafting, brainstorming, learning.
Not fit for: Acting on Gmail, Calendar, Notion. Routines. Memory of your week.
Several products will let you connect Gmail, a calendar, and a few tools, then take simple actions. Setup is usually one signup and a handful of OAuth screens. You share infrastructure, model accounts and memory layers with thousands of other customers. It is fast to start and, for most people, the cheapest way to put an AI agent on top of your existing accounts.
Fit for: Trying agent workflows quickly, low-sensitivity inboxes and calendars, hobby use.
Not fit for: Workloads where you want to be the only person whose data sits in the agent — multi-tenant by definition is not that.
A named AI agent with its own server, its own memory, and the ability to act on your tools after one tap. It lives in Telegram today (WhatsApp and iMessage in private beta). Setup is hands-off: you pick a name and a region, and we provision a dedicated agent in your name. Memory stays yours; sensitive actions wait for approval before anything is sent; you can disconnect or delete in one tap.
Fit for: People who want an agent that is theirs, in their messenger, with consented context, routines, and approval before anything sensitive is sent.
Not fit for: People who do not want a messenger-native agent or who prefer a public, shared chatbot tab.
We are not the fastest way to put an AI surface in front of you — a chatbot tab is. We are the easy, safe way to get an agent that is yours, in the messenger you already keep open: consented context, approval before anything sensitive is sent, disconnect or delete in one tap.
A general chatbot like ChatGPT or Claude takes about a minute to sign up for. That is the fastest path to an AI surface — but it is a chatbot, not an agent. The quickest path to an agent-shaped product is usually a multi-tenant assistant (signup plus a few OAuth screens) on shared infrastructure. Ermes is a different trade: it is the easy, safe way to get an AI agent that is yours — your own server, your own memory, consented context, and approval before anything sensitive is sent.
Minutes to fill in your name, pick a region and a messenger, then hours while we provision your dedicated agent in the background. You do not open a terminal. You do not configure software. The point of the easy path is that you spend minutes; we spend the hours.
Easy means no servers, no provider tokens, no terminal — you pick a name and a region, we set up the rest. Safe means consented context (you choose what the agent can use), approval before anything sensitive is sent, and a disconnect-and-delete control in one tap. The agent runs on a server in your name, not on shared infrastructure.
Three things: deciding which tools to connect, deciding what the agent is allowed to do without asking, and (if you self-host) configuring a server. Ermes removes the third. The first two we walk you through during onboarding so you ship with a working trust boundary instead of an empty one.
Yes. Telegram is the first-class messenger for Ermes and is live today. You talk to your agent in a normal Telegram thread — text, voice notes, forwarded messages. It can stage actions there and ask for one tap to approve.
WhatsApp is in private beta. The shape is the same as Telegram: voice notes, forwarded messages, one-tap approvals for anything sensitive. If WhatsApp is your default, mention it on the beta form and we will pair the cohorts accordingly.
iMessage is in private beta alongside WhatsApp. Signal, Discord and Slack are on the roadmap in that order. We add a messenger to the agent rather than asking you to leave your messenger for the agent.
Because the messenger is where your week already happens. You read texts walking around. You record voice notes mid-task. You forward messages without thinking. An agent that lives there is reachable in five seconds; an agent that lives in another tab is reachable when you remember it exists.
In the common-sense sense of "agent": not really. ChatGPT writes the answer. An AI agent finishes the task on your tools. ChatGPT has scheduled tasks and connectors in beta, but the centre of gravity is the chat tab. Ermes is the other shape — the messenger, the routine, the staged action with one-tap approval.
You can route Ermes through your ChatGPT Plus / Pro / Business subscription in priority beta, where the plan is eligible. You keep paying ChatGPT once. Ermes adds the messenger, the memory, the routines and the approval-gated actions on top.
Three things, mostly: keep a memory of your week that you can prune and export; run routines you approve once and let recur; and stage actions on your tools — send the email, book the meeting, file the note — that wait for one tap to fire.
Yours. Each Ermes account is provisioned on its own server in the region you pick. There is no shared database that mixes you with another customer’s prompts. No shared model account either — we hold zero-retention agreements with model providers, and you can bring your own key if you would rather route through your own.
Only what you give it. The first session is a short consented setup — what you want it to know about, which messengers and tools to connect, which routines you want. You can prune any memory item, export everything as JSON, or wipe the agent and start over.
No, by default. Your server is isolated and staff do not have standing access. If you ask for support you can grant time-boxed, read-only access that auto-expires in 24 hours. Every staff session is logged and visible to you.
No. Anything that touches the outside world — emails, calendar invites, charges, files leaving your account — stages itself first. The agent shows you the draft in the messenger, and one tap fires it. You can require approval on absolutely everything, or relax that scope per tool over time.
Any permission, any tool, any time. One tap revokes; the agent stops being able to use that tool in the same heartbeat. There is also a global pause that freezes every routine until you unpause.
You see what it staged before it sent. You can edit, skip, or undo. The architecture assumes the model will sometimes be wrong — that is why nothing irreversible runs without your tap.
A chatbot with connectors still lives in a tab and still forgets between conversations. Ermes lives in the messenger you already keep open, remembers your week, runs routines on its own server, and stages action — so the model is the same shape; the surface and the trust boundary are not.
Easy setup, safe by default, your own dedicated agent. Tell us who you are and what you'd ask your agent in the first week.
Request a private beta invite