Get started
What MyHermes is
What you get with MyHermes, how your own agent runs, and where you can talk to it.
MyHermes gives you Hermes, a personal AI assistant. Anyone with an account can chat with it in the browser, and a paid plan gives you your own agent on its own machine, which you can also reach on Telegram and which can run tasks for you on a schedule.
What you get
There's a free tier and two paid plans, Mini and Pro.
- Free. Sign in and chat with Hermes in your browser. Free chat runs on a shared MyHermes AI endpoint, so you don't get an agent of your own, and Telegram and backups aren't included. Free chats don't carry over: there's no Recents list, and once you close the tab you can't reopen the conversation. Hermes can also do one thing for you every day, such as a morning briefing, and you read the results in For You or by email.
- Mini and Pro. Your own dedicated agent on its own machine. Mini's machine has 2 vCPU and 4 GB of RAM; Pro's has 8 vCPU and 16 GB. A paid agent can talk to you on Telegram, remember you between chats, run tasks on a schedule and connect to other apps you use.
Prices, limits and what each plan includes are on Plans and credits.
How your agent runs
A machine of its own
Each paid agent runs on its own machine with its own disk. In builder mode, SettingsStatus shows the machine's size, for example "2 vCPU · 4 GB RAM".
On the free tier there's no machine to show: the same page says "Free chat is running on shared infrastructure."
Sleeping and waking
When your agent has nothing to do, its machine can stop. That's expected. The dashboard still counts the agent as ready, and the next message wakes it up.
- In the web chat, your message waits while the agent wakes, for up to 3 minutes. If it still isn't up after that, the chat says "Hermes is taking longer than usual to respond (180s). Try sending again in a moment." or "Network hiccup. Try again."
- On Telegram, the bot sends a short holding line such as "Let me think on that…" first, and then the real reply.
- For scheduled tasks, MyHermes wakes the agent shortly before each task is due, even if it's asleep.
- Finished replies stay put. A reply that already came in still shows after you refresh the page, even if the agent has gone back to sleep.
You can restart your agent yourself: press "restart" on SettingsStatus, or send /restart as a message on its own in the web chat or on Telegram. Hermes replies that it'll be back in about 30 seconds.
Your data
- Memory. By default your agent keeps its memory as plain Markdown files (
MEMORY.mdandUSER.md) on its own disk. No third-party account is involved. - Keys and secrets (model keys, your Telegram bot token, tool keys, webhook signing secrets) are encrypted (AES-256-GCM) before they're stored.
The Privacy Policy has the full picture.
Where you talk to it
The web chat
Everyone gets the web chat, in the dashboard's "Chat". Type in the "Message Hermes…" box and press Enter to send (Shift+Enter starts a new line).
On a paid plan, your conversations are listed under "Recents" in the sidebar. Each one is marked by where it came from: a globe for web chats, a Telegram icon for Telegram chats, and a clock for scheduled runs. Scheduled runs stay hidden until you click "Show scheduled runs". On the free tier there's no Recents list. To start a fresh conversation, click "Chat" in the sidebar on a computer, or "+ New Chat" in the top bar on a phone or a narrow window.
Telegram
You create your own bot with Telegram's @BotFather and connect it under SettingsChannels. Only the bot's owner (the first person to send it /start) and anyone whose Telegram id you add can talk to it. Messages from anyone else never reach your agent, which is why you send /start yourself as soon as you've connected it. The Telegram guide walks through it.
Your own code and other services
- An API endpoint. SettingsAPI gives your agent an OpenAI-compatible address and a key, so your own code (or any app that speaks the OpenAI format) can send it messages. See API endpoints.
- Webhooks. SettingsWebhooks gives you a web address that wakes your agent when another service sends it a signed request. The agent works on the request and posts its reply to your Telegram chat. See Webhooks.
What it does on its own
Scheduled tasks
You ask for a scheduled task in plain words in the chat, for example "send me a morning brief at 7", and Hermes sets it up. There's no form to fill in. To change or stop a task, you ask in the chat too.
The "Scheduled" page (in builder mode) lists every task with its schedule and when it runs next. In simple mode, "For You" in the sidebar shows what Hermes is doing for you on a schedule. On the free tier, For You is where your daily task and its results live.
Results go to your Telegram chat. A task you created in the web chat is sent to your Telegram home channel, which is set when you connect Telegram. Without Telegram the result isn't sent anywhere, and the row says "Not sent to any chat." That note only appears while your agent is awake, so a row without it isn't proof the result was sent. You can still read each run from the row's "See latest result" link or under "Recent runs". Each scheduled task can run up to 24 times a day on Mini (about once an hour) and 96 times a day on Pro (about every 15 minutes).
Memory
A paid agent remembers what you tell it between chats. The built-in memory needs no setup. In builder mode, SettingsMemory lets you switch to a dedicated memory service instead (Honcho, Mem0, Supermemory or Hindsight), using your own key for it.
Tools, connectors and skills
When Hermes uses a tool to answer you, the chat shows a row with a 🔧 and the number of tool calls above the reply. Open it to see each call.
In builder mode you can give it more to work with:
- Connectors (SettingsConnectors) link Hermes to apps you already use, such as Notion, GitHub, Slack or Linear. Most connect by signing in to that app, some ask you to paste a key, and a few need neither (Connectors lists them all).
- Skills (SettingsSkills) each teach Hermes a tool. Every paid agent comes with a built-in skill called Jev, for quick yes/no, pick-one and scoring questions. While your agent runs on MyHermes AI, Jev is paid from your AI credits; on another provider it needs an OpenRouter key (Skills explains).
- Tool keys (SettingsTool keys) let Hermes search the web and drive a browser with your own keys for services like Firecrawl, Tavily, Exa, Parallel, Browserbase and Browser Use.
- Local servers (SettingsLocal servers) are extra tools that run inside your agent. You add or remove them by asking in the chat. Mini can have 4 active at once and Pro 12 (Connectors explains).
The model behind it
A new agent starts on MyHermes AI, the built-in model service, so you don't need a key from anyone. What it uses is paid from your AI credits, and every paid plan includes some each month. In builder mode, SettingsModel lets you switch to another provider (OpenRouter, OpenAI, Anthropic, NVIDIA NIM and more than thirty others) with your own key, or sign in with your ChatGPT or Grok account. Paid users can also switch model from a button next to the chat box.
Simple mode and builder mode
The dashboard has two layouts. Simple mode shows the everyday parts; builder mode shows everything.
Accounts created after the start of 1 September 2026 (UTC) begin in simple mode. Older accounts begin in builder mode. Either way, you can switch whenever you like.
| Simple mode | Builder mode | |
|---|---|---|
| Sidebar | Chat, For You | Chat, Apps, Workflows, Scheduled |
| Settings | Status, Theme, Channels, Billing, Usage, Credits, Help | All of those, plus Voice, Memory, Background tasks, Model, Tool keys, Personality, Skills, Connectors, Local servers, Webhooks, API and Backups |
In builder mode, the full Settings list is grouped under "Agent", "Models", "Customize", "Account" and "Developer", with Help at the end. The billing card also shows your machine's resources.
Switching modes
- Open the account menu at the bottom of the sidebar and choose "Settings". It opens on "Status".
- Click "Builder mode: off" to turn builder mode on (it then reads "Builder mode: on"). Click it again to go back.
The sidebar and the Settings list change straight away, and the choice is saved to your account. If the switch fails, you'll see a short message asking you to try again.
Builder mode works on the free tier too, but most of what it adds needs a paid plan. Those sections show a "Pick a plan" button instead. The switch isn't shown while a new paid agent is still being set up.
Finding your way around
- The account menu. Your account row sits at the bottom of the sidebar, with your picture, your name and your plan. Its menu includes "Settings", "Usage", "Get help", "Plan & billing" and "Sign out". On a phone or a narrow window, the bar at the bottom of the screen also has "Settings".
- Settings opens as a window over the page you're on; its close button is labelled "Close settings". Each section has its own link. If you try to leave with unsaved changes, it asks "Discard unsaved changes?" first.
- Changes that need a restart. Some settings (such as Voice, Memory and the skill curator switch on Background tasks) are saved first and applied when the agent restarts. After you close Settings, a banner at the top of the dashboard offers "Apply changes". Messaging pauses during the restart, and the banner says "Your agent is back and ready." when it's done. The time zone on Background tasks is different: its "Apply" restarts your agent straight away.
- Hiding the sidebar. On a computer, "Hide sidebar" folds the sidebar away and "Show sidebar" brings it back.
- While a new agent is set up, every Settings section except Theme, Billing and Help says "Setting up your agent…" until it's ready.
Next steps
- Quickstart: create your account, pick a plan, connect Telegram and set up your first scheduled task.
- Plans and credits: prices, the free tier, AI credits and fair use.
- The Chat, Telegram, Scheduled tasks and Builder mode guides go deeper on each part.
- The docs home lists everything.