When Monad hits an error

Sometimes Monad stops partway through a reply and leaves a short red line in the thread instead of an answer. Most of the time this resolves itself and you don't need to do anything. This page tells you which is which.

Most errors fix themselves#

Monad talks to an AI model provider over the network, and that connection occasionally fails — the provider gets slow, drops the response halfway, or is briefly overloaded. When that happens Monad tries again automatically: first with the same provider, then a different one, then a different model if it has one available.

You'll see a line like:

The model provider stopped responding partway through the answer. Monad retries this automatically; if nothing happens for a few minutes, send your message again.

That message means what it says, in both halves. The retry happens in the background and you'll usually get your answer a few moments later — so you don't need to do anything yet, and re-sending immediately just means you may end up with two answers.

When to give it a nudge#

Retries aren't unlimited. If a thread has been sitting on an error for a few minutes with no new activity, the automatic attempts are used up — send your message again. That's what the second half of the message is telling you, and it's why the wording never promises that a particular error will resolve itself.

When an agent refuses to hand work to another agent#

Agents can pull other agents in — mentioning one, or handing off a focused piece of work. There are limits on how far that can go, so a chain of agents calling agents can't run away with your workspace's budget. When one of those limits is reached, you'll see a line explaining which, and it isn't an error: Monad stopped on purpose.

Every one of these messages says whether waiting helps, so read that part first.

"retry in …" or "wait up to …" is a spacing or rate limit — two agents handed work back and forth faster than allowed, too much agent work started at once, or too many runs going at the same time. Give it the stated time and ask again.

"waiting will not change that" or "does not refill" is a hard limit for this particular chain of handoffs: it has gone as many levels deep as it may (chain depth limit), this pair of agents has already worked together once here (repeat handoff), or the chain has used its whole run budget. Waiting won't help — but starting over will. A new message from a person begins a fresh chain with every one of these limits reset, so re-asking in your own words is usually all it takes. You can also just ask the agent you're already talking to to do the work itself, which is what the message's "do this work here" is suggesting to the agent.

If you hit these during ordinary work rather than something unusual, tell an admin — how deep an agent's handoffs may go is configurable per agent in Settings → Agents.

When to tell an admin#

A few errors mean your workspace needs an actual change, and retrying won't help:

• Anything mentioning credit or billing. Monad's AI usage runs on a balance, and when it runs out every agent stops until someone tops it up. The message will say whether it's your workspace's own AI key or the shared Monad balance.
• Anything mentioning a key or authentication. If your workspace connects its own AI provider account, that key may have been revoked or rotated.
• The same error repeating across different threads and different people. One person hitting one error is usually transient. Everyone hitting the same one is not.

Workspace owners and admins can check the AI settings and the recent run history in Settings → Agents, which shows each run's status and what it was working on.

Seeing what Monad was doing#

Every run leaves a trail. Open the agent's activity in the thread to see the steps it took — which tools it used, what it looked up, and where it stopped. If you're reporting a problem to your admin, the thread link plus roughly when it happened is enough for them to find the run.