When OpenClaw stops working on Hostinger, the cause is almost always one of four things: your AI credits ran out, the messaging connection dropped, the model call is failing, or (on the VPS route) the container itself is down. Each has a documented fix, and most take under five minutes.
Work through this page top to bottom: the checks are ordered from most common to least common, so the first sections solve the majority of cases. If you have not set OpenClaw up yet, start with the setup guide instead.
Fix 1: “Out of credits” / agent suddenly stops thinking
The single most common “OpenClaw stopped working” report is simply an empty credit balance. The agent replies normally while credits last, then goes silent or returns a credit error when they run out — which feels like a breakdown but is just billing.
- Check your balance in the hPanel OpenClaw/AI section. Managed plans show nexos.ai credits (for models) and Oxylabs credits (for web search) separately — you may be out of one but not the other.
- Top up from the dashboard. Credits apply immediately; no restart needed.
- Find the leak. If credits vanished faster than expected, check the heartbeat setting: the agent wakes on a schedule (default every 30 minutes) and each wake-up burns model credits even when idle. Lengthen the interval (
agents.defaults.heartbeat.every) or assign the heartbeat a lighter, cheaper model. - Check which model is active. Frontier models cost several times more per million tokens than light ones (nexos.ai publishes per-model rates). An agent silently switched to a heavy model explains most “my credits evaporated” mysteries.
Fix 2: WhatsApp / Telegram disconnected
Messaging bridges drop for mundane reasons: the phone lost network, the WhatsApp linked-device session expired, or the bot token was revoked. Symptoms: you message the agent and nothing comes back, while the dashboard shows the agent as healthy.
- WhatsApp: open WhatsApp > Settings > Linked Devices and confirm the OpenClaw session is still listed and active. If it is gone or expired, re-scan the QR code from the Hostinger dashboard.
- Telegram: confirm the bot token is still valid in BotFather. If you regenerated the token or deleted the bot, paste the new token into the OpenClaw messaging configuration.
- After re-pairing, send a test message and confirm the reply arrives in the correct chat before assuming anything else is broken.
- One channel at a time when debugging. If you run several messaging bridges, disable the others while you fix one — cross-talk between bridges is a real source of confusion.
Fix 3: Model errors (API failures, wrong key, rate limits)
If the agent connects but every request fails with a model or API error, the problem sits between OpenClaw and the model provider:
- Managed + nexos.ai credits: confirm you have a credit balance (Fix 1) and that the selected model is currently available. If one model errors persistently, switch to another model temporarily — provider-side incidents happen.
- Your own API key (either route): verify the key is still valid in the provider’s dashboard, that it has budget/quota remaining, and that it was pasted without extra whitespace. A surprising number of “model broken” cases are a key that expired or hit its spending cap.
- Rate limits: rapid-fire requests (or a tight heartbeat loop) can trip provider rate limits. The error message usually says so explicitly; the fix is to slow down, not to reconfigure.
- OAuth (ChatGPT Plus/Pro): if you connected a ChatGPT subscription via OAuth instead of credits, re-authorize when the session expires — the symptom is identical to an invalid API key.
Fix 4: Dashboard / interface will not load (managed)
If the OpenClaw web interface itself will not open from hPanel:
- Hard-refresh the browser page and try a different browser to rule out a local caching issue.
- Check whether hPanel itself loads — if Hostinger’s panel is having issues, the app interface goes with it. See is Hostinger down for how to check.
- If the app shows a provisioning or error state that persists beyond a few minutes, that is a Hostinger-side issue: contact their support via hPanel chat rather than debugging your configuration.
Fix 5: Container down or unhealthy (VPS route)
On the self-hosted route, “OpenClaw not working” can mean the Docker container stopped. SSH in and run docker ps:
- Container missing: it stopped. Check
docker logsfor the crash reason — out-of-memory kills are the classic cause on small plans. - Container restarting in a loop: read the logs; usually a bad config value or an unreachable dependency. Fix the config, recreate the container.
- After a VPS reboot: if OpenClaw never comes back after maintenance, the container lacks a restart policy. Add
restart: unless-stoppedand recreate it. - Disk full: Docker logs and old images can fill a small VPS disk over months.
docker system prune(carefully) reclaims space.
Fix 6: Agent behaves strangely (not a crash, but wrong)
Sometimes nothing is “broken” — the agent just started doing odd things. The documented culprits:
- Too many skills enabled. Each skill changes the agent’s behavior. Disable recently added skills one by one until the behavior normalizes.
- Corrupted memory/context. If the agent’s long-term memory picked up something wrong, correct it explicitly in chat or reset the memory for that topic.
- Model downgrade side effects. Switching to a cheaper model can reduce reasoning quality noticeably on complex tasks. If weirdness started right after a model change, that is your answer.
- Stale pairing. A half-expired WhatsApp session can deliver messages out of order or drop them. Re-pair cleanly (Fix 2).
When to contact support vs fix it yourself
Fix it yourself: credit balance, messaging re-pairing, API key issues, model selection, heartbeat tuning, VPS container restarts — everything on this page. Contact Hostinger support: the managed app shows a persistent provisioning error, the dashboard will not load while hPanel works fine, or billing/credit purchases do not apply. Check OpenClaw’s own channels: suspected bugs in the open-source software itself (the OpenClaw Foundation’s docs and community).
FAQ
Why did my OpenClaw agent suddenly stop responding?
The most common cause is an empty AI credit balance — check it in hPanel first. Next most common: the WhatsApp/Telegram session expired and needs re-pairing. Actual software crashes are rarer than these two.
How do I check my nexos.ai credit balance?
In hPanel, in the OpenClaw/AI section of the dashboard. Model credits (nexos.ai) and web-search credits (Oxylabs) are tracked separately, so check both if only some features stopped working.
Why are my credits draining so fast?
Usually the heartbeat (scheduled wake-ups, default every 30 minutes) combined with a heavy model. Lengthen the heartbeat interval or move the heartbeat to a lighter model; also confirm no runaway scheduled task is firing repeatedly.
OpenClaw works in the dashboard but not on WhatsApp — why?
The messaging bridge is separate from the agent. Re-check the linked-device session in WhatsApp > Settings > Linked Devices, or the bot token for Telegram, and re-pair from the Hostinger dashboard.
Do I lose my agent’s memory if I reinstall or switch apps?
On the managed plan, switching between OpenClaw, Hermes Agent, n8n, and Paperclip starts the new app fresh — configurations and history do not transfer. Export anything important before switching.
Is there a status page to check if Hostinger itself is down?
Yes — see our Hostinger status check guide for how to verify whether an outage is on Hostinger’s side or yours before you start debugging.
Reading OpenClaw logs: where they live and what errors mean
When the dashboard gives you a generic error, the logs give you the truth. Log locations differ by route:
- Managed plan: check the hPanel OpenClaw section for an activity or error log view. Hostinger surfaces the most recent errors here, though with less detail than raw container logs.
- VPS route: SSH in and run docker logs against the OpenClaw container. Add --tail 200 to see the last 200 lines, or -f to follow in real time while you reproduce the problem.
Common log signatures and their meanings:
| What you see | What it means | Fix |
|---|---|---|
| 401 / unauthorized from model API | API key invalid, expired, or over quota | Regenerate the key at the provider, update config, restart |
| 429 / rate limit exceeded | Too many requests too fast | Slow the heartbeat, add delays between scheduled tasks |
| Out of memory / OOMKilled | Container exceeded its RAM allocation | Move to a larger VPS plan or reduce concurrent workloads |
| ECONNREFUSED on messaging bridge | Bridge process crashed or lost its session | Re-pair the channel per Fix 2 on this page |
| Credit / quota exhausted messages | nexos.ai or Oxylabs balance at zero | Top up in the dashboard; no restart needed |
Slow responses and timeouts: the performance fixes
Not every problem is a crash. Sometimes the agent works but takes minutes to respond, or times out on tasks it used to handle. The causes, in order of likelihood:
- Model latency, not your server. Frontier models are slower than light ones. If you recently switched to a heavier model, the slowdown is expected. Switch back for a speed test before blaming infrastructure.
- Context bloat. Long conversation histories and large memory files get sent with every request, inflating both latency and cost. Periodically summarize or prune the agent's memory. A fresh session that responds instantly while the old one crawls confirms this diagnosis.
- Tool call chains. An agent that searches ten web pages, reads five files, and runs three commands before answering will be slow no matter what. This is working as designed. Reduce the scope of the request or accept the latency.
- VPS resource contention. On small plans, OpenClaw competes with the OS and any other containers for CPU. Run docker stats during a slow response. If CPU is pinned at 100%, the plan is undersized for the workload.
- Network egress issues. If the agent is slow specifically on web search or external API calls, the VPS network path may be congested. This is rare but real; try a different data center region if it persists.
Update and upgrade problems
OpenClaw updates can break working setups. Defensive habits:
- Snapshot before updating on VPS. Take a Hostinger VPS snapshot in hPanel before pulling a new OpenClaw image. If the update breaks something, roll back in minutes instead of debugging for hours.
- Read the changelog first. Breaking changes to configuration format, environment variables, or skill APIs are announced in releases. Five minutes of reading prevents the most common post-update failures.
- Pin versions in production. Use a specific image tag in your Docker compose file rather than latest. Update deliberately on your schedule, not whenever a new release drops.
- Managed plan updates are Hostinger's job. If the managed app breaks after a Hostinger-side update, that is a support ticket, not a DIY fix. Report it with the error text and the time it started.
Multi-instance and multi-app conflicts
Running OpenClaw alongside other AI apps (Hermes, n8n) on the same account or VPS creates specific conflicts:
- Shared credit pool confusion. On the managed plan, all apps draw from one nexos.ai balance. If OpenClaw stops but Hermes works, check whether another app drained the shared pool. The AI credits guide explains the shared-balance behavior.
- Port collisions on VPS. Each app needs its own ports. If OpenClaw will not start after you deployed something else, check for port conflicts with docker ps and remap before assuming a deeper problem.
- Messaging bridge contention. Two agents connected to the same WhatsApp number or Telegram bot will fight over the session. One channel, one agent. If you run multiple agents, give each its own number or bot token.
- Switching apps wipes config. On the managed plan, switching from OpenClaw to another app starts fresh. This is documented, but people still lose work to it. Export skills, memory, and configuration before switching, every time.