How to Fix n8n AI Assistant Blocked When AI Credits Hit Zero

Introduction

You open the Assistant, type a prompt, and instead of a workflow you get a refusal. Your workflows still run, but the chat builder is dead. This is not a bug. Since n8n 2.44.0, the Assistant stops working the moment your AI credits hit zero, and stays stopped until the balance comes back.

Quick answer

Top up Assistant credits in the Assistant credits tab of the n8n Cloud admin dashboard, or wait for the monthly reset on the 1st. Self-hosted instances use their own model provider, so this block does not apply to them.

What broke, and why

n8n 2.44.0 shipped on October 13, 2026, and its changelog explains this error in one line: “core: Block the assistant when AI credits hit zero (#40659)”. The n8n 2.44.0 release notes and the pull request behind it confirm it. Before this version, an empty credit pool produced only a generic “Something went wrong…” message, while GET /rest/instance-ai/credits returned a 500. The 2.44.0 change turns that confusing failure into an explicit block.

The block lands on a credit system n8n changed on October 1, 2026, when the Assistant moved to a newer model. As n8n announced when top-ups launched, “if you’re on n8n Cloud (Starter or Pro) and you run out of Assistant credits, you can buy more and keep building.” Starter includes 1,600 Assistant credits a month, Pro 4,000 or 9,600 depending on tier, and trials get a one-time 800. n8n’s rough guide prices the work: a quick question costs 10 to 20 credits, a simple fix 50 to 100, and a bigger build or long debugging session 300 to 600. One heavy debugging afternoon can burn through a Starter month.

Running a workflow never uses Assistant credits, so executions keep working while the Assistant is down. Only the chat builder is gated.

The fix

On n8n Cloud, check your balance first. In the Cloud admin dashboard, the Assistant credits tab shows your balance split into included and extra credits, a spend chart, and top-up history. The editor’s Assistant panel also shows the conversation’s credit use and warns you at 10 percent.

Then top up. Buy a credit package manually, or turn on auto top-up with a trigger balance, a refill target, and a monthly limit that is on by default. Included credits are spent first, and bought credits carry over and expire 12 months after purchase. Only the instance owner can top up, and topping up requires an active paid plan. Full details are in the n8n announcement of Assistant credit top-ups.

If you would rather not pay, wait for the monthly reset on the 1st, and build workflows by hand in the meantime. If you are planning an upgrade to 2.44.0, our how to update n8n self-hosted guide covers the version checks to run first.

On a self-hosted instance, this block does not apply the same way: the Assistant runs on a model provider you configure, and 2.44.0 also fixed Kimi K3 and other OpenAI-compatible providers there. If the Assistant is dead, check the configured model provider, not a credit balance.

Pitfalls that make it look like the fix failed

The most common mistake is topping up the wrong balance. Assistant credits and Gateway credits are separate pools: Gateway credits pay for model calls inside your workflows, and buying them does nothing for a blocked Assistant.

A second pitfall is trying to top up as a member: only the instance owner can buy credits, so ask the owner to top up or turn on auto top-up.

A third pitfall is the trial wall: top-ups are not available during a trial, so a trial user who burns through the one-time 800 credits must upgrade first.

A fourth pitfall is the auto top-up monthly limit: it is on by default, and once it is hit, the Assistant blocks again mid-month.

How to verify the fix worked

Send one message to the Assistant and confirm it answers instead of refusing. Then open the Assistant credits tab and confirm the balance is above zero, with the new top-up listed in the history.

When this article does not apply

This fix covers the 2.44.0 block on a zero Assistant credit balance. It does not apply to self-hosted instances, where the Assistant uses your own model provider, or when your workflows fail, since workflow model calls draw on Gateway credits or your own provider keys. It also does not apply if the Assistant fails while your balance is still above zero: that points at a server-side problem, and n8n’s community support asks for your instance URL and the timestamp of a failed attempt. Before 2.44.0, the same empty balance showed up as the generic “Something went wrong” message.

Conclusion

The blocked Assistant is a billing gate, not a bug: n8n 2.44.0 made it explicit so the cause is visible. Check the Assistant credits tab, top up or wait for the 1st, and watch the 10 percent warning so the next heavy debugging session does not stop mid-build. Our n8n complete guide covers the rest of the production checklist.

Frequently asked questions