Bring your
agent
to life. Give it the work.
A private agent in your messenger. Prompt it like a teammate: it knows your context, can take action, and asks before anything sensitive.
Free for 7 days · no card- Lives in your messenger — Telegram today, WhatsApp next
- Talk or type — voice notes in your language
- Asks before anything sensitive
Already paying for ChatGPT? Bring that plan to your agent.
Ermes is the private agent layer around the AI plan you already trust: memory, routines, approvals, and actions in your messenger. If your ChatGPT plan is eligible, we can wire your agent to use it instead of asking you to pay for a separate AI seat.
- ChatGPT Plus / Pro / Business · bring the plan you already use, where supported
- Ermes adds the agent layer · memory, reminders, approvals, messenger, and actions
- Managed AI or API keys still work · use our default setup, or connect your own provider
Availability depends on your plan, provider limits, and early-access setup. We keep Ermes's managed model account as the default, then let you flip to your own plan when you're ready.
A persistent presence — not another chat tab.
Three things your AI agent quietly does for you, in the messenger you already keep open. Nothing autonomous without your say-so.
Keeps track of what you said you'd do.
Without a separate app, calendar plugin or sync conflict.
Writes before you're ready to send.
Notes, replies, decisions — in your voice, kept in your messenger.
Calendars, repos, inboxes — with explicit permission.
Sensitive actions stage themselves and ask before they happen.
A chatbot writes the answer. Your agent finishes the task.
Two of these answer questions. One of them is connected to your calendar, your inbox and your messenger — and authorised to act once you nod.
Forget typing. Just say it.
Walking to the train, driving, between meetings, mid-run — send a voice message in your messenger. Your AI agent listens, understands, and either does the thing or asks one clarifying question. No app to open. No keyboard.
Grows with you. Not around you.
Your AI agent doesn't ship finished. It learns what you do, adjusts to your routines, nudges you when you ask it to, and slowly turns into the one tool that actually knows how you work — and what gets you going.
An agent with a real second brain.
Most AI chats forget you the moment you close them. Ermes runs a real memory on your own private server — and organizes it the way Andrej Karpathy describes building a personal knowledge system: keep every raw memory, then distill the important parts into living notes that stay current as your life changes.
See what your agent knows
The Memory Atlas shows your agent's memory as a living orb — every memory a star placed by meaning, not by app. Watch recalls fire between related memories like synapses, see new ones wire themselves in, search by idea and the map lights up. Click any star and it tells you why it's there — which conversations used it, and what it connects to.
Live for every agent today.
Notes that rewrite themselves
Beyond raw memory, your agent keeps living pages — one per person, project, and topic in your life — and rewrites them as things change, the pattern Karpathy calls an LLM-maintained wiki. Every page shows exactly which memories it was built from, and when you ask your agent to forget a source, everything derived from it is erased too.
Rolling out now.
Reaches out — without being annoying.
Your AI agent doesn't only respond. It pings you when something matters: a morning brief, a pattern it noticed, an idea you'd want, permission to do the thing you'd otherwise miss. You set the cadence. You can pause it any time. Nothing pings unless you'd thank it for pinging.
Good morning. Today is mostly focus — the partnership call at 16:00 is the only thing that needs your full head. I left two notes at the top of the Q3 doc; the one starred is the one I'd fix first.
I noticed Joon's birthday is in 4 days — and you didn't reply to his last message. Want me to draft something warm now, in your voice, so it lands on Thursday morning?
You've replied to 7 'urgent' threads this morning and haven't touched the deep-work block. The block is yours to keep or move — what would you like to do?
Routines for how you work. Ask any of them by name.
Tap any routine to see exactly what your agent does, what it needs, and what it asks before acting — in use cases.
- Idea captureVoice note → tagged idea
- Drafting in your voicePosts, replies, scripts — your phrasing
- Audience reply triageThe DMs that deserve a real answer
- Publishing checklistThumbnail, schedule, cross-post
- Newsletter recapPulls from your week, in your shape
- Brand & collab memoryEvery conversation you have had with them
- Voice reminders"Remind me when I get home"
- Family group memoryNames, kids, what they care about
- Travel planningFlight, hotel, school-run, all in one
- Health & gym nudgesOn your real cadence, not a default
- Bills, renewals & adminDrafts the boring letter, flags the form
- Weekend & errand planWhat to do, what to buy, in what order
Your private AI agent, alive in your messenger.
Most of the tools you already use, in one calm conversation. Some are built in; for the rest you sign in once.
Do it yourself, or let us do it.
Same private AI agent, two paths. The short version of what changes when we run the infrastructure for you.
The things people actually ask. Answered plainly.
What does my agent remember?
People, projects, your tone, decisions you have made. Editable and exportable, never opaque.
Can I delete or export my memory?
Yes. Prune any item, export everything, or wipe the agent — in your settings.
Does it act on my tools without asking?
No. Sensitive actions stage themselves and ask for one tap to approve. You can require approval on everything.
Where does my agent live?
On its own server in the region you pick — EU, US or APAC. Not a shared multi-tenant pool.
What messengers are supported?
Telegram is live; WhatsApp is coming soon and iMessage is planned. Signal, Discord and Slack are next.
Can I run my agent on a subscription I already pay for?
Yes — in early access. If your ChatGPT Plus / Pro / Business plan is eligible, Ermes can use it as the AI layer while adding memory, routines, approvals, messenger access, and actions. You can also use Ermes’s managed model setup or connect your own provider/API key.
Is Ermes affiliated with OpenAI or ChatGPT?
No. ChatGPT is a trademark of OpenAI. Ermes is independent and is not affiliated with, endorsed by, or sponsored by OpenAI. ChatGPT plans, limits, and billing are managed by OpenAI.
How is this different from ChatGPT?
A chatbot writes the answer; your agent finishes the task — in the messenger you already keep open.
What is a Routine?
A repeatable workflow your agent runs by name — "Friday review", "morning brief". You approve each one before it can run.
What if my agent gets something wrong?
You see what it staged before it sends. You can edit, skip or undo. Mistakes never become irreversible without your tap.
Create the AI agent you'll actually talk to every day.
Three minutes to set up. One name to give it. Your messenger already open.