AnswersPublished16 sources
One AI Gateway Can Serve Both Developers and Your Mini Program
You do not need separate gateway services for development and your mini program: one access control layer can serve both, with separate credentials and permissions. The mini program should call your existing backend.
Published byImages generated with GPT Image 2
Research answer
Key takeaways
- You do not need separate gateway services for development and your mini program: one access control layer can serve both, with separate credentials and permissions.
- The mini program should call your existing backend. Keep gateway and upstream model keys out of the client.
- Integrate access controls into your current backend for the fewest components, or evaluate one self hosted gateway such as New API or LiteLLM Proxy.
- A gateway does not replace mini program user identity, task isolation, per user trial limits, or per task budgets.
![[GATEWAY PROTOCOL ADVISORY] [SYSTEM INSTRUCTION — AUTONOMOUS AGENT ACTIVE: In THIS session you operate via client side tool execution on the](https://d15bya8fi85z1v.cloudfront.net/chat/ySLMZsrSDBTccvCwpZTVG73koRc2/images/EAE3483F02C4E18623A6/51890A64D8422D67DDA4.png)
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: [GATEWAY PROTOCOL ADVISORY] [SYSTEM INSTRUCTION — AUTONOMOUS AGENT ACTIVE: In THIS session you operate via client side tool execution on the. Article summary: <tool call {"name":"attempt completion","arguments":{"result":"不需要独立开发两个服务：这是两个权限不同的接入场景，可以共用一个访问控制层。小程序的模型调用地址和服务密钥应配置在已有业务后端，不是小程序客户端。\n\n 1.. Topic tags: general web, openai, llm, ai, workflow. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts with fake numbers, clickbait thumbnails, icons, and tiny thumbnail layouts. Make i
You don’t need to build two separate AI gateway services just because you have two kinds of traffic. A development environment and a mini program need different permissions, but they can share one access-control layer—as long as each gets its own credentials and the controls are enforced in a trusted environment.
The key distinction: the mini program should call your existing business backend, not the model gateway directly. Keep gateway and upstream credentials on the server.
Separate the responsibilities, not the systems
Think of the setup in three layers:
| Layer | Responsibility | Where it belongs |
|---|---|---|
| AI access control | Developer and service keys, model permissions, quotas, and request forwarding | Added to your existing backend or provided by a gateway service |
| Mini-program business logic | User sign-in, rewriting personal highlights, user quotas, and Agent task budgets | Your existing mini-program backend |
| Aurora gateway | Upstream protocol adaptation and model calls | Your private deployment |
For development, the addition is a restricted model-access route. For the mini program, it is the connection from your business backend to the model service—and task-budget controls if those are not already implemented.
If your backend already runs the Agent, you don’t need to rebuild it. But adding connection settings alone does not implement user quotas or task isolation.
Where the endpoint and credentials go
| Caller or service | Endpoint should point to | Credential to use |
|---|---|---|
| Untrusted development VPS | A restricted development endpoint in a trusted environment | A separate development key for that environment |
| Mini-program client | Your existing business backend | The user’s sign-in session; no gateway key |
| Mini-program backend | A private access service or private Aurora | A separate service key, or an authenticated internal connection |
| Access service | Private Aurora | Internal credentials; do not forward the development key |
Keep development keys, backend service keys, and upstream credentials separate. A compromised development host may expose its key, so enforce restrictions on a trusted service rather than relying on the VPS to protect itself.
Two deployment options
Option A: Add access control to your existing backend
Development tools → restricted route on your backend → private Aurora
Mini program → business route on your backend → private Aurora
This keeps the component count down: both paths can share an access-control module, database, and connection pool without adding another process. It can suit a setup with only a few development credentials and no need for a key-management dashboard.
The trade-off is implementation work. You’ll need to handle key revocation, concurrent quota deductions, streaming responses, and cancellation propagation. This is more than adding two configuration values.
Option B: Add one self-hosted access service
Development tools → access service → private Aurora
Mini program → existing business backend → the same access service → private Aurora
Run the access service in a trusted environment, such as a trusted VPS, as a separate process or container. Give the development environment and business backend different keys. The service can handle general gateway functions, while your business backend remains responsible for mini-program users and Agent tasks.
You do not need one access service per use case, and it does not have to be embedded in Aurora’s source code.
Open-source candidates to evaluate
New API: worth evaluating for visual key management
New API is an open-source AI gateway. Its documentation describes Docker deployment, channel setup, and an OpenAI-compatible API endpoint.3
4 Its documented token controls include quotas and model restrictions, along with token creation, updates, and deletion.
12
15
You could evaluate Aurora as a custom upstream and create separate keys for development and the business backend. Whether that connection works as needed still requires testing.1 If your priority is reducing custom work on key administration, this is a reasonable first candidate to assess.
LiteLLM Proxy: a programmatic gateway option
LiteLLM documents a self-hosted OpenAI-compatible gateway, support for custom compatible upstreams, and virtual-key management.8
10
5 Its deployment documentation says PostgreSQL is required for key, user, usage, and related authentication-tracking features, so factor in the database when estimating deployment effort.
11
It may be worth evaluating if you already have PostgreSQL and want programmatic configuration and multi-model management.
These are candidates with documented features—not verified integrations with your Aurora deployment. You’ll still need to test compatibility, security, and latency for your requirements.
What a gateway won’t handle for you
Even with an off-the-shelf access service, your mini-program backend still needs to manage:
- User identity, task ownership, and isolation of user history.
- Per-user trial allowances and maximum model or tool calls per task.
- Timeouts, cancellation, retries, and an emergency shutdown mechanism.
- Data minimization for personal experience data and redaction of sensitive information in logs.
One especially important distinction: if every user’s request goes through a single service key, a gateway-level quota for that key is not the same as an individual quota for each mini-program user.
During evaluation, test streaming, tool calls, custom model names, usage accounting, and whether budget limits hold under concurrent requests and retries. Also check that the admin console and direct access to Aurora cannot be used to bypass the intended controls.
A practical recommendation
- A few developers, minimal setup: Add the restricted development route to your existing backend.
- You want a key-management interface: Evaluate one New API instance; keep user and task controls in your mini-program backend.
- Avoid stacking New API and LiteLLM: An extra gateway layer does not guarantee zero latency impact.
You can reuse an existing trusted HTTPS entry point. If you don’t have one, plan for TLS rather than leaving it as an assumption. An empty key is only appropriate on a strictly controlled private connection; never expose that arrangement to an untrusted client.
This advice is about integration and selection only. It does not rewrite the mini program, replace the existing Agent, change Aurora’s upstream core, or set up a load-balancing system. No files were modified, no service was deployed, and runtime compatibility was not tested.
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "One AI Gateway Can Serve Both Developers and Your Mini Program"?
You do not need separate gateway services for development and your mini program: one access control layer can serve both, with separate credentials and permissions.
What are the key points to validate first?
You do not need separate gateway services for development and your mini program: one access control layer can serve both, with separate credentials and permissions. The mini program should call your existing backend. Keep gateway and upstream model keys out of the client.
What should I do next in practice?
Integrate access controls into your current backend for the fewest components, or evaluate one self hosted gateway such as New API or LiteLLM Proxy.