跨 300 个 POP 的 WASM，
5ms冷启动。
Multi-language WASM runtime (Rust, Go, TypeScript, Python). Crons, DLQs, secrets, streaming, unified AI Gateway. Billed to the exact CPU millisecond.
Deno.serve(async (req) => {const form = await req.formData()const file = form.get('document')// OCR / WASM processingreturn Response.json({ status: 'processed', file: file?.name })})
从 HTTP 处理程序到 ML 密集型异步作业
WASM multi-runtime
Rust, Go, TypeScript, Python compiled to WASM. Cold-starts under 8ms, memory-safe, fully sandboxed.
Cron scheduler
Standard crontab + extensions (@every 15m, @weekly). Drift guaranteed under 100ms, manual triggers supported.
Job queue + DLQ
NATS JetStream under the hood. Exponential backoff, automated DLQ, one-click replay from Studio.
Streaming responses
SSE, HTTP/3 streaming, response chunks. Ideal for LLM streaming, transcoding, ML inference.
AI Gateway
Unified proxy for OpenAI, Anthropic, Mistral, Llama. Semantic caching, per-tenant budgets, observability baked-in.
Preview deployments (roadmap)
One preview URL per GitHub/GitLab PR, cloned environment variables — planned roadmap feature.
部署在 300 个 POP 中，在 12 个区域可见
您的代码在距离用户最近的 POP 上运行。 12 个核心区域托管独立的运行时，而 288 个边缘 POP 则智能地路由和缓存流量。
在生产中测量，而不是综合基准
Aurabase Edge · Cloudflare Workers · Vercel
每天 240 万个作业 · 99.99% SLA
« 每天 240 万个 WASM OCR 作业，成功率达 99.99%，成本仅为我们之前的 Kubernetes 堆栈的四分之一。仅在语义缓存方面，AI 网关就为我们节省了 6 个月的工程时间。 »
常见问题
Why WASM instead of V8 isolates?+
How does the AI Gateway work?+
What about secrets management?+
What is the maximum execution duration?+
幂等性密钥 标头。Can we deploy existing code (Deno, Bun, Express)?+
获取（请求） 处理程序按原样运行。 Express 中间件可以通过轻量级垫片进行调整。建议将完整的旧版 Node 应用程序用于 aura-compute（专用虚拟机），而不是边缘功能。