Components

Where traffic starts. A domain resolves to one host, so Users take one link.

Round robin across up to 8 hosts. No health checks: a dead host keeps its share.

Splits traffic across App Servers and drops a dead host at its next health check.

Answers cacheable pages near the user. Egress from the edge costs less.

Blocks bots before they reach the app, and some real users with them.

Runs your code. Answers requests and sends queries to the database.

Takes work now so Workers can do it later. Users wait only for the enqueue.

Runs your code on tasks from a Queue. Users never wait for it. Link it to the data an export reads.

Link App Servers or Workers into it. They pack onto shared nodes: 15% less Burn, copies in 30 s.

Serves AI prompts in tokens per second. Each size step doubles the GPUs.

The source of truth. One process per connection, 100 connections at most.

Many app connections share 20 database connections.

Answers read queries from memory while the hot data fits. Writes pass through and invalidate.

Answers reads. Lags behind writes, so some API reads come back stale.

Hashes passwords for you. Charged per user.

Answers search queries. 400 per second per size step.

Holds uploaded files. No limit. Pay for what it stores.

Charges cards for checkouts. 2.9% plus $0.30 per checkout.

Answers report scans in columns. 1.5 s each, no limit, paid per scan.

Serves AI prompts with no GPUs. Paid per token, and rate limited.

Sends push notifications and email. Paid per message, 10,000 a second.

Replicates each key to N nodes. Reads ask R of them, writes wait for W.

SQL on Raft. Every write waits for a majority of its replicas.

Revenue waits here. The drawer holds about a minute of it.

Link any Node into it. It samples that traffic and turns it into Insight.

Link App Servers or Workers into it. It adds copies when they run hot and removes them when idle.