Hostinger’s AI products — managed OpenClaw, Hermes Agent, n8n, and the AI Builder — run on AI credits instead of requiring your own OpenAI or Anthropic account. There are two separate credit types: nexos.ai credits pay for the language model doing the thinking, and Oxylabs credits pay for web search and data extraction.
This guide explains how each credit type works, what burns them, what the heartbeat costs, and how top-ups work — using only Hostinger’s published information, with no invented prices. Where Hostinger has not published a number (credit-pack prices at checkout), we say so plainly instead of guessing.
The two credit types
| nexos.ai credits | Oxylabs credits | |
|---|---|---|
| What they pay for | Language model calls — the “thinking” | Web search and data extraction |
| Used by | OpenClaw, Hermes Agent, AI Builder, n8n AI features | Agent web-search skills, research tasks |
| Included at start | Pre-configured with the managed plan (“Ready-to-use AI”) | Free starter allocation (“Instant web search”) |
| Alternative | Your own provider API key, or ChatGPT Plus/Pro via OAuth | None — or do research tasks without web search |
The key point: the balances are separate. An agent can be fully funded for thinking but unable to browse, or vice versa. When something stops working, check both balances before assuming a bug — see OpenClaw not working.
How nexos.ai credits burn
nexos.ai is Hostinger’s model layer: instead of you holding accounts with Anthropic, OpenAI, Google, and others, Hostinger routes your agent’s model calls through nexos.ai and deducts credits per token. Hostinger’s support documentation publishes the per-model rates per million tokens — a few examples from the October 2026 price list:
- Gemini 3 Flash: 0.5 credits per 1M input tokens / 3 per 1M output
- GPT-4.1: 2.2 in / 8.8 out
- Grok 4: 3 in / 15 out
- Claude Sonnet 4.5: 3.3 in / 16.5 out
- Claude Opus 4.6: 5.5 in / 27.5 out
- Claude Opus 4: 15 in / 75 out
Two things drive your burn rate: which model (a 30x spread between the cheapest and priciest above) and how much it talks (output tokens cost 4–6x input tokens on every model). A chatty agent on a frontier model burns credits an order of magnitude faster than a terse agent on a light model doing the same job.
Practical translation: routine work — summaries, inbox triage, scheduled checks — belongs on a light model. Reserve the frontier models for work where reasoning quality visibly matters: complex coding, nuanced writing, multi-step planning. Most users who complain about credit burn are running everything on the most expensive model out of habit, not need.
The heartbeat: the silent credit drain
OpenClaw and Hermes Agent can wake themselves on a schedule — the heartbeat, every 30 minutes by default — to check whether anything needs doing. Every wake-up runs the model, which consumes nexos.ai credits even when there is nothing to do.
Do the rough math: 48 wake-ups a day, each burning a small but nonzero amount of tokens. Over a month that is ~1,440 model calls doing nothing but checking. For an always-on personal assistant this is the price of proactivity; for an agent you only use interactively, it is pure waste.
Three documented ways to control it:
- Lengthen the interval. The heartbeat cadence is configurable (
agents.defaults.heartbeat.everyin the gateway settings). Hourly or every few hours is plenty for most personal use. - Use a lighter model for the heartbeat. The check-in does not need frontier reasoning — a cheap model deciding “nothing to do” costs a fraction.
- Turn it off if your agent is purely reactive (only acts when you message it). Zero heartbeat, zero idle burn.
How Oxylabs credits burn
Whenever your agent searches the web or extracts data from pages — “check the price of X,” “summarize this article,” scheduled web monitoring — it spends Oxylabs credits rather than nexos.ai credits. The managed plan includes a free starter allocation (“Instant web search,” 1,000 Oxylabs AI Studio credits per Hostinger’s checkout). Heavy research agents and website monitors consume these steadily; a chat-only assistant barely touches them.
If web features stop working while chat still works, your Oxylabs balance — not your nexos.ai balance — is the empty one. They are independent pools.
Top-ups: how to buy more
Both credit types are topped up from the Hostinger dashboard — no separate nexos.ai or Oxylabs account needed. Credits apply immediately to your balance. One honest caveat: Hostinger’s support articles mention adding credit packs at checkout (e.g., “up to 100 nexos.ai credits”) but do not publish the USD price of those packs publicly, so we do not print one here. Check the live price in your hPanel cart before buying — that is the only number that matters.
Your alternatives to buying credits:
- Your own API key. Paste an OpenAI/Anthropic/Gemini key and pay the provider directly per token, skipping nexos.ai entirely. Worth it if you already have provider billing or negotiated rates.
- ChatGPT Plus/Pro via OAuth. Connect a ChatGPT subscription in hPanel for flat-rate model access with no per-token billing — attractive for heavy users, since subscription pricing does not scale with usage.
- Local models (VPS route). Self-hosted OpenClaw on a VPS can point at a local Ollama instance: zero per-token cost, at the price of weaker models and your own GPU/RAM.
How to keep credit costs under control
- Match the model to the task. Light models for routine work, frontier models where quality visibly matters. This one habit dominates all other savings.
- Tame the heartbeat. Longer interval, cheaper model, or off — as above.
- Watch the balance weekly at first. Check both credit pools for the first month until you know your burn pattern; surprises only happen to people who never look.
- Keep prompts tight. Long system prompts and huge conversation histories inflate input tokens on every call. Prune context you do not need.
- Batch scheduled work. One daily briefing run costs less than the same checks scattered across 48 heartbeat wake-ups.
Worked example: what do credits actually buy?
Credit math is abstract until you attach it to real usage. Using Hostinger’s published per-model rates, here is what a fixed amount of thinking power looks like on two different models. Take a hypothetical 100 nexos.ai credits (the checkout add-on quantity Hostinger names) and spend them entirely on output tokens, the expensive kind:
- On Gemini 3 Flash (3 credits per 1M output tokens), 100 credits buys roughly 33 million output tokens — an enormous amount of chat, summaries, and routine agent work.
- On Claude Opus 4 (75 credits per 1M output tokens), the same 100 credits buys roughly 1.3 million output tokens — plenty for serious work, but about 25x less raw volume than the light model.
This is not a pricing claim — it is arithmetic on Hostinger’s published rates, and it shows why model choice dominates your credit bill more than any other decision. A useful rule of thumb: if you cannot articulate why a task needs a frontier model, it probably does not.
For input tokens the picture is even friendlier, since input rates run 4–6x cheaper than output on every model. Long conversation histories and big pasted documents cost input tokens; the agent’s actual replies cost output tokens. When people report surprisingly high burn, the culprit is usually long outputs on a heavy model — verbose answers on Opus rather than concise ones on Flash.
Do credits expire? What happens at renewal?
Credit balances are tied to usage, not to your billing term — topping up adds to the pool you draw from as the agent works. Your plan renewal (the hosting subscription) and your credit balance are separate things: renewing the plan does not refill credits, and running out of credits does not cancel the plan. Treat them as two wallets: one pays Hostinger for hosting, the other pays for intelligence.
FAQ
What are nexos.ai credits?
nexos.ai is Hostinger’s model layer for its AI products. Credits pay per token for the language model your agent uses — you do not need your own OpenAI or Anthropic account. Rates vary by model and are published per million tokens in Hostinger’s support docs.
What are Oxylabs credits for?
Web search and data extraction by your agent — a separate balance from nexos.ai model credits. The managed plan includes a free starter allocation; if web features stop while chat works, this is the empty pool.
Why do my credits drain when I am not using the agent?
The heartbeat: scheduled self-wake-ups (default every 30 minutes) that run the model each time, even with nothing to do. Lengthen the interval, use a cheaper model for it, or disable it for purely reactive agents.
How much do credit top-up packs cost?
Hostinger does not publish credit-pack USD prices publicly — check the live price in your hPanel cart before buying. We do not print a number we cannot verify.
Can I use my own API key instead of credits?
Yes. On the managed plan you can paste your own provider API key at onboarding (skipping “Ready-to-use AI”), and on the VPS route it is the standard setup. You can also connect a ChatGPT Plus/Pro subscription via OAuth for flat-rate access.
Do unused credits expire?
Credits are a usage pool separate from your hosting subscription term — plan renewal does not refill them, and empty credits do not cancel the plan. Check Hostinger’s current terms in hPanel for the latest expiry policy.