Berlin, Germany · Est. 2021
data infrastructure that actually works at the edge
The first thing you need to know about Upstash is that they charge you for using their product. Not for having it. This sounds obvious until you realize most database providers do the opposite. You provision an instance, you pay for it whether you send one query or a million. Upstash charges per request. You send nothing, you pay nothing.
Here's the thing about serverless functions. They pop into existence, do something, and disappear. Sometimes in under 50 milliseconds. Databases want to stay open. These two things don't get along.
Traditional Redis providers charge by the instance. You pick a size, you pay monthly. The connection stays open. Works fine for a server that runs 24/7. Does not work at all when your code lives inside a Cloudflare Worker that spins up and dies in the time it takes to sneeze.
Establishing a TCP connection to Redis takes longer than some edge functions live. So you either hold connections open (can't do that in edge runtimes) or eat the overhead on every request (kills your performance). Either way, you pay for an instance that sits idle because serverless traffic is spiky by nature.
Enes Akar and Çağatay Çalı kept running into this wall while building serverless applications. Their solution was straightforward, even if it sounded wrong at first: make Redis talk HTTP. Build a REST API that accepts standard Redis commands over HTTP, so any environment that can make a web request can use Redis. No TCP needed. No connection pooling. Just a POST request with your command, get a response, done.
A Redis command that never arrives because you can't connect is slower than a Redis command that arrives over HTTP. Zero is slower than something.
The initial reception was mixed. Developers who actually worked with edge functions immediately got it. Redis purists were less enthusiastic - turning Redis into a REST service felt like a dilution. HTTP adds overhead. Everyone knows that.
What the purists missed: in edge runtimes, the alternative to Upstash's REST API wasn't fast TCP connections. It was no Redis at all. Over two years, as edge computing went from niche to mainstream, "no Redis" stopped being acceptable. Upstash's approach went from weird to obvious. Multiple Redis providers have since added HTTP endpoints of their own.
Handles product and business. Background in distributed systems. What makes him unusual: he still thinks like an engineer. Per-request pricing, REST-first, generous free tier - reads like someone who was personally annoyed by the alternatives, not someone who ran a competitive analysis spreadsheet.
Leads go-to-market, which mostly means: write good docs, write good blog posts, let developers find Upstash.
Built the technical core. Architected the REST proxy that translates Redis protocol into HTTP-friendly APIs. This is harder than it sounds. Redis was designed around a binary protocol that assumes a persistent connection. Teasing that apart into stateless HTTP requests while maintaining full command compatibility requires understanding both protocols at a deep level.
Made Redis work in places where it physically could not run before.
The original. Serverless Redis with REST API, per-request pricing, global replication. Still the most-used Upstash product and the default Redis option in the Vercel ecosystem.
Message queue that delivers to any URL via POST. Retries, delays, scheduling, fan-out. Built because the Upstash team needed it internally. Then realized everyone had the same gap.
Serverless Apache Kafka. Managed clusters behind a REST API. Kafka is notoriously complex to run - Upstash hides all of that behind per-request pricing.
Vector database for AI/LLM apps. HNSW indexing, metadata filtering, REST API. Prototype a RAG pipeline on Upstash without committing to a monthly minimum.
Rate limiting as a service. Sliding window and fixed window algorithms. Send a request, get a boolean. No infra, no algorithm to implement.
Workflow engine for durable orchestrations in serverless. Functions time out on long processes. Upstash Tempo handles state management so yours don't have to.
Instead of stitching together Redis Cloud + AWS SQS + Confluent Kafka + Pinecone, you get all four from Upstash. One provider, one pricing model, one SDK.
Vercel
Cloudflare
Deno Deploy
AWS Lambda
Roughly two to three dozen people, distributed globally with a European anchor. No enterprise sales team. No expensive conference sponsorships. No "schedule a demo" forms. You find Upstash through the blog, the Vercel Marketplace, or another developer. The product speaks. Developers who like it tell other developers.
The Upstash blog is one of the higher-quality technical blogs in developer tools. Not repurposed press releases - deep technical explanations of specific problems. These posts regularly surface on Hacker News, which became one of the company's main growth channels. The blog is marketing, except it's actually useful.
They built QStash because they needed a message queue internally. Couldn't find one that worked in serverless. Built it. Looked around. Realized every serverless developer had the same gap. Shipped it as a product.
Cache reads at the edge. Your Cloudflare Worker or Vercel Edge Function checks Upstash Redis for cached data without establishing a TCP connection. Low latency because of regional deployments. Low cost because per-request pricing.
Build a RAG pipeline without a monthly commitment. Upstash Vector stores embeddings and runs similarity searches via REST. Prototype on the free tier (10K requests/day). Scale linearly. No minimum spend.
Queue background jobs from serverless. Upstash QStash lets your edge function push a message, then return immediately. Work happens later, delivered to any URL as a POST. Trivial with a standard backend, nearly impossible in serverless - unless you have QStash.
Rate-limit an API without managing infra. Upstash Ratelimit: send a request, get a boolean. Sliding window or fixed window. No Redis to set up, no algorithm to implement, no state to manage.
Run Kafka consumers in serverless. Upstash Kafka: managed clusters over REST. Produce to or consume from topics without managing connections or consumer groups. Kafka without the operational burden.