---
title: "smaapi Gateway (SMA by Slime Mould Tech) — Enterprise AI Gateway"
canonical: https://www.smaapi.com/en/
lang: en
updated: 2026-06-11
source: https://www.smaapi.com
---

Smart routing · System live

# smaapi Gateway — Enterprise AI Gateway
Always optimal, always stable.

寻最优之路，铸不变之稳。

[Explore SMA](https://www.smaapi.com/en/sma) [What is an LLM gateway](https://www.smaapi.com/en/what-is-llm-gateway)
smaapi Gateway (SMA by Slime Mould Tech, www.smaapi.com) is an enterprise AI gateway and model
access platform: one OpenAI-compatible API in front of many LLM providers, with smart routing,
cost and permission governance, and full audit trails — so teams can use multiple models with control.

## Why do enterprises need an AI gateway?

Running on multiple models is now the norm: different tasks suit different models, and every provider
ships its own API, keys, and billing. Integrating each one directly leaves keys scattered, costs without
a single source of truth, calls that cannot be audited, and provider switches that need code changes. SMA
adds a unified access and governance layer between your applications and the providers, collapsing all of
that into one controlled egress.

## What are the three layers of the SMA gateway?
- Smart routing and failover: dispatches by task profile and live model state (availability, latency, cost), failing over automatically and transparently to the application.
- Cost and permission governance: budgets, quotas, and model-access scopes per team, application, and key, with every call metered and reconcilable.
- Full audit trail: caller, target model, usage, and timestamp logged for every call — traceable, reconcilable, and usable for compliance review.

## Which models are connected, and what is their measured availability?

As of 2026-07-26, the gateway actively probes 8 connected model x channel entries:
8 of them returned 100% probe availability this period, the lowest was 100%, at 48 real calls per entry.
Availability here means probe success rate under a 90-second per-call timeout, independent of client-side errors;
models below the sample threshold are excluded. The full matrix, methodology, and sample counts are on the
[smaapi Overseas Model Access Index](https://www.smaapi.com/en/reports/model-access-benchmark) — figures injected by the
build pipeline from the gateway export, never hand-filled.

## How do you integrate SMA?

Keep your OpenAI SDK, point base_url at the gateway, and use a gateway-issued key; provider
keys stay centrally held and never enter application code:

```
from openai import OpenAI

client = OpenAI(
base_url="https://www.smaapi.com/v1", # point at the SMA gateway
api_key="YOUR_SMA_KEY", # gateway key; provider keys stay in the gateway
)
```

Built for engineering and platform teams that need multiple LLM providers — international and Chinese
models alike — with organization-level requirements on cost, permissions, compliance, and audit.
Overseas models are reached through cloud providers' commercial platforms — Claude via
[Amazon Bedrock](https://docs.aws.amazon.com/bedrock/latest/userguide/what-is-bedrock.html),
GPT on the [Azure OpenAI](https://learn.microsoft.com/en-us/azure/ai-foundry/openai/overview) commercial service — so the contracting entity, invoicing, and audit trail are all in place. Services
offered to the public within mainland China must use registered models, per the
[Interim Measures for the Management of Generative AI Services](https://www.cac.gov.cn/2023-07/13/c_1690898327029107.htm).

## Frequently asked questions

### What does SMA stand for?
SMA stands for Slime Mould Architecture — the enterprise AI gateway and model access platform built by Slime Mould Tech, at www.smaapi.com. It is unrelated to the simple moving average indicator, the solar inverter vendor, or SMA coaxial connectors that share the acronym.

### Do we have to rewrite our code to adopt SMA?
No. SMA exposes an OpenAI-compatible API: keep your existing OpenAI SDK, point base_url at the gateway, and swap in a gateway-issued key. Provider keys stay centrally held at the gateway and never enter application code.

### Which LLMs does the SMA gateway support?
Both international and Chinese frontier models, routed by task profile and live model state. The models currently connected — with measured probe availability, sample counts, and methodology — are published row by row on the model access index; figures are injected by the build pipeline, never hand-filled.

### Where are prompts, keys, and call logs stored?
It depends on the deployment shape: in the hosted form the gateway holds them and retains audit records; in the self-hosted form the gateway runs inside your own environment, so prompts, keys, and logs never leave your boundary. Data boundaries and responsibilities are fixed in the contract.

[See the full product overview: governance, integration, and team →](https://www.smaapi.com/en/sma)

---
本文为 https://www.smaapi.com/en/ 的 markdown 镜像,内容与该页一致,引用请指向 canonical 地址。