when something breaks

no product with a real machine behind it is failure-free, and a page that pretends otherwise is the first thing you would distrust after your first bad afternoon. this is the honest version: what usually goes wrong, what to try, in order, and what to do when none of it works.

it is taking a long time

give it a minute first. your agent does real work on a real machine, so a request that sends it to read three websites or write a file takes longer than a chat window would. if you want to know what it is doing, open the live desktop and look.

if it has been quiet for several minutes on something small, send the message again. if that does not land, open the web dashboard and ask there; the dashboard is the channel to trust when you need to know whether a message arrived, because you can see the conversation on the page.

it says something went wrong

send /new. that starts a fresh session and clears whatever the last one was stuck on. your agent keeps its memory of you; only the thread resets. if the same request fails twice in a fresh session, the problem is the request or the site it depends on, and your agent should be telling you which.

it stopped for the day

that is the daily allowance, not a fault. every plan has one, and when it is used up your agent pauses until it resets the next day and tells you so. plans and billing explains the numbers.

it seems to have forgotten something

ask it directly what it has written down about the thing:

what do you have in your notes about my flight on the 14th?

if the note is there, the problem was the question. if it is not, tell it again and it will write it down now. memory explains what it keeps and how to see it.

a message never arrived on iMessage

this is the one we cannot see. we do not get delivery receipts back from the iMessage carrier, so a message that never reaches your phone is invisible to us until you tell us. send again, or use the dashboard. it is not common, and it is real.

it was unreachable for a few seconds

once a day, at a quiet hour, your agent's software restarts itself to stay healthy. the restart takes around fifteen seconds. a message sent during that window is picked up when it comes back, which is almost immediately; you will rarely notice, and now you know why if you do.

i want to start completely over

the dashboard's reset agent memory wipes its conversations, what it learned about you and the identity it built, and keeps your wallet, channels, plan and skills. there is no undo. privacy and your data has the exact list.

my own telegram bot stopped answering

if you run your agent under your own bot rather than the shared one, the token from BotFather is the thing to check: it has to be pasted whole, including the colon in the middle, or the connection fails silently. revoke the token in BotFather, create a new one, and paste that. channels explains the two ways to use Telegram.

what can you see from your side?

the dashboard shows your agent's health: healthy, or not. when it is not, the platform restarts a stuck agent on its own; you do not have to ask. if the dashboard keeps showing it unhealthy, that is the moment to write to us, and the health status is the first thing we will look at too.

none of that worked

email help@instaclaw.io from the address on your account, or post in the support channel on our Discord at instaclaw.io/discord. say which channel you were using and roughly when. people who run InstaClaw can reach your agent's machine to fix it, which is why a support message can end in a fix rather than a shrug, and it is also why privacy and your data tells you that plainly.

it answered from a different model

if the model your agent usually uses is unavailable, your agent does not go quiet: the reply comes from another model, and it says so at the top of the message, naming both, inside the platform notice frame that is never written by your agent. you are told which model answered every time it happens; a swap you were not told about is the one thing this page rules out.

what this page does not promise

a response time for support, or that every failure has a fix on our side. some do not: a website that refuses automated visitors, a service that is down, a phone that is out of signal. your agent is built to say which of those it is, and this page is built so you know what to try before you ask.

what next

• talking to your agent: what it does when it cannot
• channels: which channel to trust when a message seems lost
• memory: checking what it actually wrote down
• how your agent thinks: the tiers, the picker, and the swap