Help

Troubleshooting

What a slow, silent or failing agent usually means, what to check, and how to fix it.

Most problems show up as a message in the chat, a note on a Settings page, or a bot that stays quiet. Find what you're seeing below: each entry says what it usually means, what to check and how to fix it, and links to the page that covers it in full.

Quick checks

A few things are worth a look before anything else:

Is your agent still being set up? While a new paid agent is set up, the message box is disabled and most Settings sections say "Setting up your agent…". Wait until that's gone.
Is it running? SettingsStatus (open it) says "Agent is live" when your agent is running or asleep, and both are normal. "Starting up your agent…" means it's still being set up. "Agent is offline" means neither; press "start" to try bringing it up. When it's live, "restart" restarts it, and so does sending /restart as a message on its own.
Is there a banner at the top of the dashboard? Some settings only apply after a restart. Close Settings, and if a banner offers "Apply changes", press it.
Is your plan active? SettingsBilling shows your plan, or "No active plan".

The web chat

The first answer takes a while

Your agent was asleep and has to wake up first. That's normal, and you don't need to do anything: the chat box stays usable, and the reply arrives once your agent is up. The chat waits up to 3 minutes.

If your agent still isn't up by then, you see either "Hermes is taking longer than usual to respond (180s). Try sending again in a moment." followed by a short detail, or "Network hiccup. Try again." Either way, send your message again.

Chat explains sleeping and waking.

Hermes didn't respond in time

The chat says Hermes didn't respond in time and that your model provider may be slow, not responding, or out of credits. This mostly happens on a message with an image attached: it gets up to about 4 minutes once your agent is awake (less if it had to wake first), and the reply took longer than that.

Check your balance. If your agent runs on MyHermes AI, open SettingsCredits (open it) and top up if it's low. With your own provider key, check your balance with that provider.
Then try again in a moment.

Usage and credits shows how to top up.

A reply takes many minutes, or times out

A message without an image keeps running in the background while Hermes works through it, and the reply appears when it's done, even after several minutes. If it still hasn't finished 15 minutes after you sent it, the chat marks it as timed out within the next 5 minutes. Send the message again; for a very long request, try splitting it into smaller ones.

Chat lists the chat's limits.

Hermes is busy, was interrupted, or lost contact

The chat says	What to do
Hermes is busy with another request	Wait a moment, then send your message again.
Hermes was interrupted while working	Something stopped your agent's work on the reply, such as a restart. Send the message again.
It lost contact with Hermes while it was working	Send the message again.

The provider rejected your key, ran out of credit or rate-limited you

These messages come from the model provider your agent uses. They point you to your model settings, which live in SettingsModel (open it).

The chat says	What to do
Your API key was rejected	Paste a fresh key from your provider into "API key" and press "Save credentials". MyHermes checks it with the provider before saving. On MyHermes AI there's no key box: keep it selected in `SettingsModel` and press "Save credentials", and MyHermes checks its own key and issues a new one if needed. If you signed in with ChatGPT or Grok, there's no key box either: press "reconnect" on the card and sign in again.
Your key doesn't have access to this model	Pick a model your key can use, or check your account in the provider's own dashboard. If you signed in with ChatGPT or Grok, pick a model from the model button next to the chat box.
Your model id isn't valid for this provider	With your key in place, pick a model from the list under "Model" instead of typing one, then save. If you signed in with ChatGPT or Grok, pick one from the model button next to the chat box instead.
Your provider is out of credits or quota	Add credit with that provider, or switch to another provider. If your agent runs on MyHermes AI, top up under `SettingsCredits`.
Your provider rate-limited you	Wait a minute and send again, or switch to another provider.
Your provider returned a server error	Try again in a moment, or switch to another provider.

Models and keys walks through switching provider.

Hermes didn't return a reply

The chat says Hermes didn't return a reply. Usually the model provider had an error: a rate limit, no credits left, or a key it won't accept. Check SettingsModel, and your balance, then send again.

Chat lists the other messages a provider can send back.

The chat says no model provider is set up

The chat opens SettingsModel with a note that no model provider is configured. Pick a provider, paste your key if it needs one, and press "Save credentials". MyHermes AI needs no key, and picking it fills in its model.

Models and keys describes the built-in option.

Hermes is catching its breath

The chat says Hermes has been hard at work and is catching its breath, and tells you when it'll be back at full speed. Your agent has reached its plan's running-time cap.

Your web chat message wasn't saved. Send it again after that time.
Telegram messages wait. They're answered automatically once the window frees up.
SettingsUsage shows a "RECHARGING" banner with when Hermes is back online.

Plans and credits explains the cap and its numbers.

"Too many messages this minute. Wait 60 seconds and retry."

The web chat takes up to 10 messages a minute for each agent. Wait a minute, then send again.

Chat lists the web chat's limits.

The message box is greyed out

The box says your agent is starting up or restarting. A new agent is still being set up, or your agent is restarting after a settings change. Try again in a few seconds.

To follow a restart, close Settings: the banner at the top of the dashboard says your agent is restarting, then "Your agent is back and ready." If the model button next to the box is greyed out instead, hover it to see why, for example "Wait for the current turn to finish".

Memory and background tasks describes that banner.

An attachment won't send

Still uploading. You can't send until every attachment has finished uploading.
Failed upload. Hover the file's chip to see why, press its × to remove it, then attach the file again.
Too big. A PDF, DOCX, XLSX or PPTX over 4.5 MB is refused. Split it into smaller files.
"This attachment is no longer available. Please re-attach the file and try again." A file you attached but didn't send within about a day has expired. Attach it again.

Chat has the file types and sizes.

The chat couldn't reach your agent or lost its connection

You see "Couldn't reach the chat service:" or "Couldn't reach your Hermes instance:" (each followed by a detail), "Network connection lost while loading chat updates." or "Network hiccup. Try again." Check your internet connection, then send your message again. "Network hiccup. Try again." can also mean your agent took too long to wake, so if it keeps happening, check SettingsStatus as described under Quick checks.

The chat says your subscription has ended

Your paid plan is over, so your agent can't chat any more. To get your own agent back, start a new plan from SettingsBilling.

Billing explains what you get back when you resubscribe.

The free tier

The free tier's limits are listed in Plans and credits.

A note says you've sent too many messages

Before you sign in, the chat takes 10 messages every 3 hours from one IP address. After that, new messages are turned away until the 3-hour window has passed. Signing up is the way to keep chatting.
On a free account with your own provider key, free chat takes up to 60 requests a minute from one IP address. Wait a minute, then send again.

"The free tier is briefly at capacity. Please try again in a moment."

This shows under a free chat reply that failed. Wait a moment, then press the small retry icon under the failed reply (hovering it shows "Retry") to send the same message again.

Other errors under a free chat reply

A free chat reply can also fail with one of these. Each shows the same retry icon under the reply.

The chat says	What to do
"The AI service is temporarily unavailable. Please try again."	Wait a moment, then press the retry icon.
"That request couldn't be processed. Please try again."	Press the retry icon. If it fails again, try again later.
"The AI service returned an error. Please try again."	Press the retry icon.
"This conversation has grown too long. Start a new chat to continue."	Retrying won't help. Start a new chat: "Chat" in the sidebar on a computer, or "+ New Chat" in the top bar on a phone or a narrow window.

Free chat on another model stops answering

Any model other than the free one, myhermes:free, is charged from your AI credit balance. When the balance runs out, replies on that model fail with an error.

Check your balance under SettingsCredits. Top up, or switch back to myhermes:free in SettingsModel and press "Save LLM settings".

Models and keys covers the free tier's model settings.

Your free daily task stopped, or its email didn't come

For You says "Paused after failed runs". The task is off and its last run didn't go through. After 5 failed runs in a row it turns itself off; it reads the same if you paused it yourself after a failed run. Press "Resume" to turn it back on; it runs again at its next daily time. Resuming doesn't reset the count, so if that run fails too, it turns itself off again. Deleting the task and setting it up again starts the count from zero.
For You says "The last run didn't go through." Hermes tries again at the task's next daily time, which the message names. There's nothing you need to do.
No email arrived. Check that "Email me each result" is ticked in For You. A run that fails sends no email, and a result whose email couldn't be sent is still in For You.
The unsubscribe page says "That didn't go through. Try again in a moment." Press "Stop these emails" again in a moment. If the page says the link looks incomplete, open the unsubscribe link from the email again.

Plans and credits covers the daily task.

A section says it's only available to paid plans

Settings sections such as Channels, Memory, Webhooks or Backups say, for example, "Channels are only available to paid plans" with a "Pick a plan" button. They configure your own agent, which comes with a paid plan.

Your free chat from earlier is gone

Free chats don't carry over. There's no Recents list on the free tier, and once you close the tab you can't reopen the conversation. A paid plan keeps your chats under "Recents".

Channels never shows connected as your bot

After saving and reloading, SettingsChannels (open it) should show "connected as @" followed by your bot's username. If it doesn't, MyHermes couldn't look up the bot with that token. Get a fresh token from @BotFather, paste it into "Bot token" and press "Save credentials".

Telegram shows what a working connection looks like.

The bot doesn't answer at all

If your bot sends nothing back:

Check that SettingsChannels shows "connected as @" and your bot's username. If it doesn't, see Channels never shows connected as your bot.
Check SettingsStatus. If it says "Agent is offline", press "start".
Give a sleeping agent a moment to wake, then send another message.
If Channels shows your bot as connected and the web chat answers, send us a ticket with your bot's username.

The bot says "I don't recognize you yet"

The bot sends this to anyone who isn't its owner or on your allowed list, and asks them to tap /start. Their messages never reach your agent.

If it says this to you, the id under "Allowed Telegram users" doesn't match your Telegram id. Check your id with @userinfobot, fix the row in SettingsChannels and press "Save credentials". A correct id there always lets you in.
If you'd already sent /start yourself in a private chat and still get this reply, someone else sent /start to your bot before you did. The first private /start makes that person the owner, and the owner can also message your agent. The dashboard has no way to change the owner. Tell us through Get help.
If it says this to someone you want to let in, add their numeric id under "Allowed Telegram users" and save.

Telegram explains who can message your bot.

The bot is slow to reply

A sleeping agent has to wake first. A short holding line such as "Let me think on that…" means your message arrived; the real reply follows once your agent is up.

The bot says Hermes is catching its breath

Your agent has reached its plan's running-time cap. The bot says so and tells you when it'll be back at full speed. Your message is saved and answered automatically once the window frees up, so there's nothing to resend.

The bot says your subscription has ended

Your plan is over, and your messages no longer reach an agent. Start a new plan from SettingsBilling to bring it back.

If resubscribing gives you a new agent, your bot, its token and your allowed ids carry over, but the owner doesn't. Send /start to your bot yourself straight away, before anyone else does.

Billing covers what changes when an agent is removed.

Scheduled tasks

A task runs at an unexpected time

Your agent runs your tasks in its own time zone. A new agent is on UTC until you set yours, so a time you type, such as "at 8am", may be saved as 8am UTC.

Check your agent's time zone in SettingsBackground tasks (open it). If the "Timezone" card says your agent runs in UTC, pick your time zone and press "Apply".
Open the Scheduled page and read the task's row.
For a repeating task, the schedule line is shown in your own time zone. If it ends with a zone name in brackets, such as "(UTC)", there's no exact match in your time zone, so it's shown in your agent's time zone.
For a one-off reminder or an "every N minutes or hours" task, the schedule line is shown as Hermes saved it, not converted. Go by "next in …" for those.
If the time is wrong, ask Hermes in the chat to change the task's time, then press "↻ refresh" and check the row again. A task saved while your agent was on UTC keeps its UTC time after you set your time zone, so it still needs this.

Hovering "next …" or "ran …" shows the exact date and time in your time zone. If your agent's time zone differs from yours, check your rows again after a daylight saving change. An app automation has its own "Timezone" setting; see App automations.

Scheduled tasks has the details.

The time zone card says your agent needs an update

"Your agent needs an update before its timezone can change." means your agent can't change time zone yet, so it stays on UTC. Go by each task's schedule line on the Scheduled page, and ask Hermes to change a task's time if it's off.

The time zone card lists some of your tasks

The card says these tasks "could not be moved" to your time zone "without changing when they run", and your agent stays in its current time zone. Ask Hermes to change or delete them, then press "Apply" again. An app automation can be in this list too. It shows as the start of its request, or as "auto-" and a number. Change its schedule or delete it on its app's Automations page.

If the card instead says your agent "could not switch" to your time zone, with no list, press "Apply" again or pick another time zone.

Scheduled tasks explains which tasks can't be moved.

The time zone card says it's taking longer than expected

"This is taking longer than expected. Try again." means the change hasn't come through after 15 minutes. Press "Apply" again. To cancel it instead, pick the time zone your agent runs in now and press "Apply".

A new automation doesn't appear

If your agent isn't on UTC, an automation whose schedule has no exact match in your agent's time zone isn't saved, and the dialog closes with no message. An edit like that isn't saved either. Open the dialog again, set its "Timezone" to your agent's time zone and save.

App automations explains the time zone rules.

A row says "Not sent to any chat."

The task's last result had nowhere to go. A task you create in the web chat sends its results to your Telegram home channel, and without Telegram there isn't one. Connect Telegram so results have somewhere to arrive.

You can still read each run from the row's "See latest result" or under "Recent runs", as long as it's among your 200 most recent chats. The note only shows while your agent is awake, so a row without it isn't proof that a result was sent.

Scheduled tasks explains where results go.

A row says it exceeds your plan's daily limit

The row says "Exceeds your plan's 24/day limit" (96 on Pro). Each scheduled task can run up to 24 times a day on Mini and 96 on Pro. MyHermes doesn't wake your agent for a task set to run more often, so it only runs if your agent happens to be awake. Ask Hermes to run it less often.

Scheduled tasks lists the limits.

A task didn't run at all

The running-time cap was reached. MyHermes doesn't wake your agent for tasks until the window frees up.
Your plan has ended. Tasks aren't woken after that.
It's over your plan's daily limit. See the entry above.
The row says "failed". The last run ended in an error. Open "See latest result" to read what happened, and ask Hermes about it in the chat.

Scheduled tasks covers the rest.

A new task isn't listed

Press "↻ refresh". Opening the page never wakes your agent, and while it's asleep the page shows a saved copy that updates when a chat starts or ends on your agent. A task set to run more often than your plan allows isn't in that saved copy at all, so it's missing until your agent is awake.

Scheduled tasks explains the saved copy.

Settings changes

A change you saved doesn't seem to apply

Voice, Memory, the skill curator switch on Background tasks, and app settings are saved straight away but only reach your agent after a restart. Close Settings and press "Apply changes" in the banner at the top of the dashboard. If an "Apply failed" message appears, the banner stays, so try again.
The time zone on Background tasks doesn't use the banner: its "Apply" restarts your agent straight away. The card shows how the change is going (see Scheduled tasks).
Personality saves restart your agent for you. Hermes reads the files when a session starts, so start a new chat to try the change. If the save message says the restart failed, press "restart" on SettingsStatus.

Memory and background tasks describes the banner, and Personality what a save does.

A card says your credentials are saved but your agent failed to start

This can follow a save in SettingsChannels or SettingsModel. Your details are saved, but they didn't reach your agent. Press "Update credentials again →", then "Save credentials" once more in a moment. On Channels, if you added ids, they're already filled in and an empty token box keeps your saved bot; if you saved only a token, paste it again.

Models and keys covers this card.

Tool keys are saved but not applied

"Saved and applied to your agent." means your agent got the keys. If the message says they're saved but not that they were applied, they didn't reach your agent. Type them in again and press "Save tool keys" in a moment.

Models and keys lists the tool services.

These come from the sign-in card in SettingsModel (open it). Models and keys walks through the sign-in.

The card says	What to do
"Code expired. Try again."	The code ran out before your approval came through. Press the button to start again and use the new code.
"Sign-in denied. Try again."	Grok reported the sign-in as declined. Start again and approve it.
"Sign-in expired or failed. Try again."	Start again and go through the sign-in once more.

After a message, the button reads "Try again: ChatGPT" or "Try again: Grok", or still "reconnect" if you were already connected. While the card says "Waiting for sign-in…", stay on the Model card until it says "connected to ChatGPT" or "connected to Grok": leaving the Model section or closing Settings stops MyHermes from picking up your approval.

Signed in, but your agent wasn't updated

The card says you're signed in but your agent couldn't be updated. Your sign-in is saved, but it didn't reach your agent. Sign in again with the card's button ("reconnect" or "Try again").

A ChatGPT model doesn't answer

Pick a model from the model button next to the chat box. With ChatGPT, it lists the models your ChatGPT account offers. Settings has no model box for ChatGPT or Grok.

Connectors

A connector shows connected, but your agent can't use it

"● connected" is the state MyHermes saved when you connected. It isn't checked again each time your agent uses the app, so a connection can look fine while its calls fail.

Press "Disconnect" on the card.
Press "Connect" and sign in again, or paste the key again.
Press "Restart Hermes" in the notice.

Then start a new chat and ask again, mentioning the app by name.

Connectors explains why this happens.

A connector shows error or awaiting auth

"● error" means MyHermes couldn't finish the sign-in, or couldn't renew access later. Press "Reconnect" and sign in again.
"● awaiting auth" means a sign-in was started but never finished. Press "Retry connect".

Connectors lists every status.

Connecting fails

After signing in on the app's site, you come back to Connectors with a message that the connection failed. Press "Connect" and try again.
With an API key, the dialog closes and a message asks you to check the key. The app turned it down, so nothing was saved. Check that you made the key where the dialog's hint says, then paste it again.

Your agent doesn't use a connector you just added

Press "Restart Hermes" in the notice that appears after you connect. MyHermes tries to hand the change to your agent by itself, but that doesn't always get through.

Connectors covers the restart step.

Webhooks

A webhook request gets an error

Response	Usual cause
`401` with `bad_signature`	A wrong secret (use the whole secret, `whsec_` included), the wrong text signed (sign the timestamp's value, a dot, then the body), a signature that isn't lowercase hex, or a body that changed after signing
`401` with `stale_timestamp`	The timestamp is in seconds instead of milliseconds, or your clock is more than 5 minutes off
`401` with `missing_signature` or `missing_timestamp`	One of the two headers is missing
`404` with `webhook_not_found`	The URL is incomplete, or the webhook was deleted
`404` with `instance_not_found`	The agent this webhook belongs to has been removed
`429`	Too many requests in a minute, or the webhook's daily limit is used up. Wait `retryAfterMs` milliseconds.

Webhooks has every response.

A webhook returns 200 but nothing arrives in Telegram

A 200 means the request was accepted, not that your agent finished.

Check Telegram is connected under SettingsChannels. Without it, your agent runs but its reply isn't sent anywhere.
Check your plan is active. After a plan ends, requests can still be accepted, but your agent doesn't run.
Check the webhook's fire count went up in SettingsWebhooks. If it did, your agent may have failed, sent an empty reply, or taken longer than 110 seconds. Try a shorter request.

Webhooks explains where replies go.

GitHub or Stripe can't call your webhook

They don't send MyHermes's two signing headers, so every request is turned away. Put a signing step in between, such as a code step in Zapier or n8n.

Webhooks shows how.

The API endpoint

An API request gets an error

Response	Usual cause
`401`	No key, a wrong key, a key that's been rotated, or an agent that has been removed. Copy the key from `SettingsAPI` again.
`403`	Your plan has ended.
`404` with `not_found`	The path isn't under your Base URL's `/v1`.
`429` with `quota_exceeded`	The running-time cap is reached. Wait until the time in `resetAt`.
`503` with `instance_warming_up`	Your agent was asleep and didn't wake within 180 seconds. Send the request again.
`503` or `502` with another error	MyHermes couldn't check your plan or reach your agent just then. Try again.

A request that runs longer than 300 seconds, counting the time to wake your agent, is cut off.

API endpoints lists every error body.

One request ran your agent more than once

The official OpenAI library for Python retries a request after a 429, a server error or a timeout, and a retry can run your agent again. Set max_retries=0 on the client, as the example in Using an OpenAI SDK does.

Backups and restores

A backup fails

"Too many requests. Wait a moment and try again." You can make up to 3 backups an hour, in an hour that starts with your first one. Wait, then try again.
Any other error under "Create backup now": try again in a moment. If your agent was asleep, it's woken first, which can take a little while.

Backups covers making backups.

A restore failed

Check your agent before doing anything else:

Send your agent a message and wait for its reply, so you know it's awake.
Open SettingsPersonality and see whether its files read as they did before the restore. If the editor shows "Last synced", you're looking at a stored copy, not your agent's disk: wait a moment and open it again. "No SOUL.md on disk yet" for a file you had before means the disk was wiped.

If it's as it was, the restore stopped before changing anything. Don't restore a pre_restore copy.
If it isn't, close and reopen SettingsBackups so History refreshes, then restore the pre_restore copy whose time matches your attempt. That puts your agent's own files back as they were just before it, and you can try the restore again if you still want it; if it fails the same way again, send us a ticket. An older pre_restore would take you further back.

Backups explains what a restore does.

An app stopped working after a restore

A restore wipes your agent's disk before putting the backup back, and a backup doesn't hold installed apps. Their files and data are gone, even though the Apps page may still list them. The pre_restore copy doesn't hold them either.

Backups lists what a backup holds.

No automatic backup today

Automatic backups are a Pro feature and stay off until you pick "daily" or "weekly". They run once a day at 05:00 UTC, and only when your newest backup of any kind is at least a day old ("daily") or 7 days old ("weekly"). A backup you made by hand pushes the next automatic one back.

Backups has the schedule rules.

Billing

Billing says a payment failed

A renewal payment didn't go through. Billing asks you to update your payment method and shows a "See plans" button. Your agent keeps working for 7 days after the failed payment. If it's still unpaid after that, your plan ends, and your agent is later removed without a final backup. For help with a payment, send us a ticket.

Billing has the timeline.

Your plan card has no buttons

Some subscriptions can't be changed from the dashboard, and their card says to contact support. Send us a ticket.

Billing covers these plans.

Still stuck

Send us a ticket from SettingsHelp, which "Get help" in the account menu opens (open it). Get help says what to put in it so we can help quickly.