The two credit systems
Every Hostinger AI product that uses AI credits — managed OpenClaw, managed Hermes, AI features in n8n — draws from the same two pools:
- nexos.ai credits pay for language-model usage, routed through the Hostinger AI Router. One key gives you Claude, Gemini, GPT-class models, Grok, and others — no separate API accounts needed.
- Oxylabs credits pay for web search and data extraction — the agent’s ability to look things up and scrape pages. These are separate from model credits and were never interchangeable.
The balance is shared across Hostinger AI products that use AI credits. Run OpenClaw and Hermes in the same month and they draw from one pool. The balance can never go below zero — when it runs low, requests are declined and agents stop generating responses until you top up. There is no overdraft and no surprise negative balance.
What the managed plan bundles
The managed AI automation plan comes with what Hostinger calls “initial AI credits” and “pre-configured AI credits” — enough to start using your agent immediately, with no API keys of your own. At checkout you can also add “Ready-to-use AI” (nexos.ai credits, up to 100, per Hostinger’s support article) and “Instant web search” (1,000 free Oxylabs AI Studio credits).
Here is the honest framing: those starter credits are a trial allowance, not a monthly quota. A casual user — a few chat sessions a week — may not touch the top-up button for months. A production user — scheduled jobs, web research, code execution — should assume top-ups are part of the real monthly cost from the start.
Example costs per model (published data)
Hostinger’s support articles publish example nexos.ai consumption rates per one million tokens. These are the closest thing to an official price list that exists:
- Gemini 3 Flash: 0.5 input / 3 output credits
- GPT-4.1: 2.2 input / 8.8 output credits
- Grok 4: 3 input / 15 output credits
- Claude Sonnet 4.5: 3.3 input / 16.5 output credits
- Claude Opus 4.6: 5.5 input / 27.5 output credits
- Claude Opus 4: 15 input / 75 output credits
Read this table as a relative guide, not a bill. It tells you that a Claude Opus 4 workload costs roughly 30x a Gemini 3 Flash workload — which makes model choice the single biggest cost lever you control. It does not tell you what a credit costs in dollars, because Hostinger had not published top-up pack USD prices as of October 9, 2026.
What we know vs what we do not know
To keep this page honest, here is the ledger:
- Known: the two credit systems and what each pays for; the shared-balance behavior; the never-below-zero rule; example per-token consumption rates per model; the checkout add-ons (up to 100 nexos.ai credits, 1,000 Oxylabs credits); where to top up (Hostinger dashboard on managed, Docker Manager → Projects on VPS).
- Not published: the USD price of additional credit top-up packs for the managed plan; the exact number of starter credits bundled with the plan; any official per-task cost examples (“a research brief costs X credits”).
If a coupon or review site prints exact top-up pack prices, treat them with suspicion unless they show a dated cart screenshot — as of our check, Hostinger had not published them. Check your dashboard or the live cart before scaling up usage.
How credits get spent (beyond your prompts)
Your prompts are not the only draw on the balance. Two background consumers surprise new users:
- Heartbeat / wake-ups: agents that wake on a schedule consume credits every time they check in, even with nothing to do. Lengthening the interval (or using a lighter model for the heartbeat) cuts this materially.
- Tool loops: web search, file reads, and code execution each cost model tokens on top of the search credits. An agent that searches ten pages to answer one question spends ten pages’ worth of reading.
The practical move: watch your first week. If the balance drops faster than expected, check whether a scheduled task is looping or a heartbeat is set too aggressively before blaming the per-token rates.
Ways to pay less (or nothing per-token)
- Bring your own key: paste your own OpenAI/Anthropic API key at onboarding and pay the model provider directly, skipping nexos.ai metering for those models.
- OAuth your ChatGPT subscription: the managed onboarding supports connecting a ChatGPT Plus/Pro subscription via OAuth — flat-rate billing replaces per-token metering for supported models.
- Default to efficient models: route routine work to Gemini Flash-class models and reserve frontier models for tasks that earn the premium.
- Trim the heartbeat: longer intervals, lighter models for wake-ups.
- Scope web search: enable it where it earns its keep; a chat-only assistant does not need scraping credits burning in the background.
Credits on the VPS route
Self-hosting does not escape the credit question — it just moves it. On a KVM VPS you can use Hostinger AI Router credits (auto-filled if purchased with the VPS), Oxylabs credits for search, or your own API keys entered manually at deploy. If credits run out or the balance would go negative, the agent stops generating responses until you top up at Docker Manager → Projects. The per-token examples above apply to the shared nexos.ai system either way.
FAQ
Two prepaid credit pools that power Hostinger’s AI products: nexos.ai credits pay for language-model usage through the Hostinger AI Router, and Oxylabs credits pay for web search and data extraction. The managed plan includes starter credits; beyond that, usage is billed via top-ups. The balance is shared across your Hostinger AI products and can never go below zero.
Hostinger had not published USD prices for top-up packs as of October 9, 2026 — which is why this page does not print one. Top-ups are bought from the Hostinger dashboard (managed) or Docker Manager → Projects (VPS). Check the live cart or your dashboard for current pack prices before scaling up.
Nothing dramatic: the balance can never go below zero, so requests are simply declined and the agent stops generating responses until you top up. No overdraft, no surprise bill. Top up from the dashboard and the agent resumes.
Yes. The nexos.ai credit balance is shared across all Hostinger AI products that use AI credits, including OpenClaw, Hermes, and AI features in n8n. Switching apps under the managed plan does not strand your balance — only app-specific configuration and history are lost on a switch.
Yes, two ways: paste your own model provider API keys (you pay the provider directly), or connect a ChatGPT Plus/Pro subscription via OAuth during managed onboarding for flat-rate billing. The starter credits just let you begin without either.
The usual suspects: an aggressive heartbeat/wake-up schedule burning credits while idle, a scheduled task looping on an ambiguous job, or a frontier model doing work a lighter model could handle. Check scheduled tasks and heartbeat settings first, then model choice — per-token rates are rarely the real culprit.
Related: Hermes pricing · automation apps catalog · AI agents hub · OpenClaw guide
Real cost scenarios: what AI usage actually costs per month
The per-token table on this page shows relative costs. Here is how those translate into realistic monthly scenarios. These use the published per-million-token rates and assume typical usage patterns. Your actual costs depend on model choice, which remains the dominant variable.
| Usage pattern | Typical monthly tokens | Model choice | Relative cost signal |
|---|---|---|---|
| Casual: a few chat sessions per week | 1-3M tokens | Mixed (Flash for routine, Sonnet for hard tasks) | Starter credits may last months |
| Regular: daily assistant use, some automation | 10-30M tokens | Mostly efficient models | Expect regular top-ups |
| Production: scheduled jobs, web research, code execution | 50-200M tokens | Mixed by task complexity | Top-ups are a line item in your budget |
| Heavy: frontier models for everything | 50M+ tokens | Claude Opus class | 30x the cost of the efficient tier for the same tokens |
The 30x figure deserves emphasis. At the published rates, one million output tokens on Claude Opus 4 costs 75 credits versus 3 for Gemini 3 Flash. Same task, same token count, 25x the credit burn. Before optimizing anything else, audit which model handles which task. Most production setups waste the most money here: a frontier model answering simple questions that a light model handles identically.
Credit monitoring: catching burn before it hurts
Hostinger's dashboard shows your nexos.ai and Oxylabs balances, but passive checking is not a strategy. Build these habits:
- Weekly balance review. Every Monday, note both balances. Calculate the weekly burn rate. At the current rate, how many weeks until empty? This single number tells you whether your usage is sustainable.
- Per-project attribution. If you run multiple agents or workflows, track which one consumes the most. The usual surprise: a forgotten scheduled task or an aggressive heartbeat burning credits while producing nothing useful. Kill or tune the top consumer first.
- Alert thresholds. Decide in advance: at what balance do you investigate? At what balance do you top up? Write these numbers down. Balance below 20% means investigate today. Below 10% means top up today. Without thresholds, you discover the empty balance when the agent stops working.
- Post-incident review. Every time credits run out unexpectedly, do a five-minute postmortem. What consumed them? Was it legitimate work or waste? Each incident teaches you something about your real usage pattern.
Oxylabs credits: what burns them fastest
Web search and data extraction credits deplete through different mechanics than model credits. The expensive patterns:
- Unbounded research loops. An agent told to research this topic thoroughly may visit dozens of pages. Each page fetch costs search credits plus the model tokens to read the content. Cap research depth explicitly: check at most 5 sources prevents runaway consumption.
- Repeated scraping of the same pages. If your workflow scrapes the same product pages or documentation daily, cache the results. Re-scraping unchanged content is pure waste. Store fetched content with timestamps and skip re-fetching within a freshness window.
- JavaScript-heavy pages. Pages requiring browser rendering for extraction cost more than simple HTML fetches. When designing scraping workflows, prefer API endpoints or static pages where available.
- Overlapping agent responsibilities. Two agents researching the same topic independently doubles the search cost. Coordinate: one agent researches, others consume the results.
Bring-your-own-key vs nexos.ai: the real comparison
The page mentions BYOK as a cost saver. Here is the fuller analysis:
| Factor | nexos.ai credits | Bring your own key |
|---|---|---|
| Setup effort | Zero; works immediately | Create accounts, generate keys, configure per model |
| Model switching | One balance, all models | Separate billing per provider |
| Cost transparency | Credit abstraction hides per-token USD | Direct per-token pricing from each provider |
| Rate limits | Shared Hostinger pool | Your own API tier and limits |
| Best for | Starting out, multi-model experimentation | High-volume single-model workloads |
The crossover point: if 80%+ of your usage is one model family (say, Claude), a direct Anthropic API key with volume pricing may beat the credit abstraction. If you experiment across models or have variable workloads, nexos.ai's unified balance is worth the opacity. Revisit this decision quarterly as your usage pattern stabilizes. And the OAuth option for ChatGPT Plus/Pro remains the best deal for individual users: flat-rate subscription billing replaces per-token metering entirely for supported models.