Zeabur AI Hub Paused? Switch to BazaarLink, One Line Change

Zeabur AI Hub Is Currently Paused

Zeabur's AI Hub is currently paused. If you used it to call models such as GPT, Claude and Gemini, you now need an entry point that can take over directly. BazaarLink can replace it seamlessly: it is also an OpenAI-compatible API, with one key for calling multiple model providers, and your code only needs a one-line base URL change. It also adds things AI Hub does not offer: per-project keys, spend caps for each key, automatic multi-model failover, and unified invoices for Taiwan companies.

This article explains how to migrate directly, and how to get key governance right while you do it.

Three Principles Before You Migrate

Principle 1: Do not put original provider API keys directly into your deployment platform. Original keys from providers such as OpenAI and Anthropic usually have no spend limit by default. As long as they are valid, an attacker can use your account to call the API without restriction until you notice an unusual bill. Keys have no built-in spending brake, so once one leaks, the only option is to stop the bleeding after the fact. The damage is already done.

Principle 2: Every project and every deployment environment should use its own independent key. If all projects share one original provider key, a leak in one environment exposes the quota of every project, and it is hard to trace an unusual bill back to the environment that caused it. Independent keys are not just a security measure; they are the foundation of cost attribution. Without them, you may not even know which service is burning money.

Principle 3: Every key should have a spend cap. This is the most overlooked defense, and also one of the most direct. Key leaks are almost impossible to rule out entirely (credentials can leak, code can be committed by mistake, third-party platforms can be breached, and any link in the chain can fail). But if each key has a daily or monthly spend cap, the maximum damage from a leak is locked to a number you can accept, instead of an "unlimited bill."

The common logic behind these three principles is: do not rely on a single line of defense. Design your system assuming that a key leak will happen, rather than hoping to avoid it by luck.

Applying These Three Principles with BazaarLink

BazaarLink (bazaarlink.ai) is an OpenAI-compatible AI API relay service. You do not need to rewrite your code: just change base_url to the BazaarLink endpoint, and you can call GPT, Claude, Gemini, Qwen, DeepSeek and other providers through the same API, without holding every provider's original key in your own system.

Step 1: Change the Base URL

For code that already calls an OpenAI-compatible API, change base_url / baseURL to the BazaarLink endpoint:

from openai import OpenAI

client = OpenAI(
    base_url="https://bazaarlink.ai/api/v1",
    api_key="sk-bl-YOUR_API_KEY",
)

completion = client.chat.completions.create(
    model="anthropic/claude-sonnet-4.6",
    messages=[{"role": "user", "content": "Summarize these meeting notes for me"}],
)

print(completion.choices[0].message.content)

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://bazaarlink.ai/api/v1",
  apiKey: "sk-bl-YOUR_API_KEY",
});

const completion = await client.chat.completions.create({
  model: "openai/gpt-5.6-luna",
  messages: [{ role: "user", content: "Summarize these meeting notes for me" }],
});

This means you do not need to hold any original provider keys. Model-side credentials are centrally managed by BazaarLink. The environment variables you deploy to Zeabur, Vercel or any other platform contain only the project key BazaarLink issued to you (starting with sk-bl-), not the original OpenAI or Anthropic keys.

Step 2: Create One Key per Project, with a Spend Cap

BazaarLink's key management API lets you name each key and set a spend cap and reset period. Creating a key requires calling the API with your account's management key:

curl https://bazaarlink.ai/api/v1/keys \
  -H "Authorization: Bearer sk-bl-YOUR_MANAGEMENT_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "side-project-staging",
    "limit": 20,
    "limit_reset": "monthly"
  }'

The newly created key is the one you put into the environment variables on Zeabur (or your other deployment platform). It is completely independent of the keys for other projects, and it can spend at most 20 US dollars per month. Even if this key does leak, the attacker's damage is locked within the limit you set, and it cannot turn into an unlimited bill.

Step 3: Turn On Automatic Failover (Currently Beta)

Another added value of a multi-model relay is failover. When the primary model's provider returns a 5xx error, times out, or is rate limited, BazaarLink can automatically retry with the fallback models you specify, without any change to your code:

{
  "model": "openai/gpt-5.6-luna",
  "messages": [{"role": "user", "content": "..."}],
  "extra_body": {
    "models": ["anthropic/claude-sonnet-4.6", "google/gemini-3.7-flash"]
  }
}

The response includes a header telling you which model actually served that request, so you can trace it afterward. This feature is currently in Beta. Before using it in production, we recommend verifying yourself that its behavior meets your service-level requirements.

A Bonus: Usage Reports

Because all model calls pass through the same relay layer, BazaarLink can break down spend and token usage by model, team, member, and API key, with export to CSV, JSON and PDF. Even without a key leak, this report is useful for monthly reconciliation and spotting unusual usage, and it is often much easier than digging through bills from several providers.

Extra Benefits for Taiwan Companies

Beyond key governance, BazaarLink is a Taiwan company: it bills in US dollars, quotes in TWD with rates updated daily, and spares you from handling cross-border card fees and exchange losses yourself. Enterprise users can have electronic unified invoices issued under their unified business number, and can also negotiate monthly enterprise billing (Net-30 payment terms, negotiable). If your company follows standard Taiwan accounting procedures, you do not need to research how to book a foreign vendor's payment receipts or deal with cross-border digital service tax issues for a single AI provider. For many small and mid-sized teams, this administrative cost is often underestimated more than the technical specifications.

Conclusion

Moving from Zeabur AI Hub to BazaarLink takes only a one-line base URL change. Use the migration to make "original provider keys never stored in deployments, one independent key per project, and a spend cap on every key" your standard architecture. From then on, no matter which platform you deploy on, key risk stays within a controllable range.

If you want to do this without changing much code, go to bazaarlink.ai and create your first project key with a spend cap, and see for yourself what it feels like to know that even if a key leaks, the maximum loss is locked down first.

TWD billing · Taiwan invoices · leading AI models · OpenAI-compatible API