Redis is the quiet infrastructure behind most small apps: session storage, rate limiting, job queues, caching, and real-time features all run through it. The good news is that you can run Redis free in 2026 in three genuinely useful ways. The bad news is that each free option has a different limit, and the limit that matters depends entirely on what your app does with it.

Upstash is the most generous serverless option. Its free tier offers pay-per-request pricing with a daily allowance that covers most small projects, plus a global replication option that puts data near your users. Because it charges per request rather than per hour, an idle app costs nothing, which makes it ideal for side projects and prototypes. The tradeoff is latency: serverless Redis adds a few milliseconds per request compared with a dedicated instance, and very chatty workloads can burn through the request allowance faster than expected.

Redis Cloud, run by the company behind Redis itself, offers a fixed free database with a 30 MB storage limit and a set number of connections. Thirty megabytes sounds small, but for sessions and small caches it is often enough, and the fixed instance gives you predictable latency and the full Redis feature set, including modules. The limit to watch is connections: a serverless frontend that opens a new connection per request can exhaust the pool quickly, so use a connection pool or a proxy in front of it.

Self-hosting is the third option, and for anyone who already runs a VPS it is usually the best one. Redis uses very little memory for small workloads: a few hundred megabytes covers most side projects, and the Docker image starts in seconds. The cost is the server you already pay for, plus the maintenance: updates, backups, and a firewall rule that keeps port 6379 closed to the internet. Never expose Redis publicly. Unauthenticated Redis instances are scanned within minutes and used for cryptomining.

Choosing between the three is a workload question. Use Upstash when your traffic is bursty or your app is serverless and you want zero maintenance. Use Redis Cloud when you want the official feature set with predictable latency and your data fits in 30 MB. Self-host when you already have a server, your data is large, or you need persistence and modules without paying per request. Many projects start on Upstash and move to self-hosted Redis once the monthly request count makes the math obvious.

One practical note applies to all three: set a memory policy before you deploy. Redis with no eviction policy will eventually run out of memory and stop accepting writes, which looks like an outage. Use allkeys-lru for a cache, so old keys are evicted automatically, and set up a simple alert on memory usage. A cache that fails gracefully is a feature; a cache that crashes your app is a bug you installed yourself.

Persistence is the detail that separates a cache from a database, and it is where Redis beginners lose data. By default, Redis keeps everything in memory, and a restart empties it. Enable AOF persistence if the data must survive a restart, or RDB snapshots if you can afford to lose a few minutes of writes. For a session store, losing sessions on restart is annoying but survivable. For a job queue, losing queued jobs is data loss, so persistence is not optional. Check the persistence setting before the first deploy, not after the first outage.