Leaderless · Coherent · Redis-compatible

Database consistency at cache speeds.

Swytch is a leaderless, Redis-compatible cache with database-level consistency. Every node accepts writes. Every node serves reads. Every node sees the same writes in the same order. Self-host it free under AGPL, or connect it to Swytch Cloud — the zero-knowledge control plane. Your existing Redis client connects unchanged.

Get it on GitHub See pricing 
# macOS / Linux (Homebrew)
brew tap swytchdb/tap
brew install swytch

# macOS / Linux (curl)
curl -fsSL https://getswytch.com/scripts/install.sh | sh

# Windows (PowerShell)
iwr -useb https://getswytch.com/scripts/install.ps1 | iex

Docker: ghcr.io/swytchdb/swytch

Pick your problem

What brought you here?

Your cache is lying to you

Every cached read inside a database transaction is data your database never validated. Watch it happen.

A cache that never lies →

Multi-region without the rewrite

Rearchitect, buy a planet-scale database, or accept eventual consistency? There’s a fourth option.

Active-active →

Some of your cache is a source of truth

Queues, streams, sessions, counters — data with no upstream to rebuild from. It deserves better than AOF.

Durability with Swytch Cloud →

Four systems doing one job

Sessions, queues, leaderboards, and cache — each with its own upgrade cycle, monitoring, and line on the bill.

Consolidate →

Your eviction policy was a guess

Someone set maxmemory-policy years ago. The workload changed; the policy didn’t. It can tune itself instead.

AI-powered caching →

Networks that disconnect on purpose

Ships, field sites, retail floors. A partition is a state your data is designed for, not an incident.

A holographic cache →

Your data stays in the EU

Dutch company, self-hosted data plane, zero-knowledge control plane. Bring the compliance team.

EU sovereignty →

How we stack up

Architectural properties, claim by claim, against Redis Cluster, CockroachDB, and Spanner. No benchmarks.

Compare →
FAQ

The obvious questions

Short answers. Longer ones live in the docs.

No primary, no leader, no coordinator. Every node accepts writes and serves reads. No leader election ever happens, because there is no leader to elect — the failure mode is gone by construction. From the application’s perspective each region acts as a single writer, and the writers merge as their histories propagate.

Yes. Swytch speaks the Redis wire protocol (RESP) on port 6379 with Redis 8 semantics. Existing client libraries connect unchanged; existing application code runs unchanged. Joining is a hostname swap.

You choose, per prefix. The default blocks the minority for strict consistency — the behavior Raft users expect. Opt in per prefix and both sides keep writing, then reconcile deterministically when the link heals: not by wall clock, not last-writer-wins, but by the order the causal histories imply.

Because self-hosting should be free, forever, with no rug to pull. AGPL keeps the open-source version genuinely open — unlimited nodes, full engine — while companies that can’t ship AGPL buy a commercial license. That’s the whole business model, in one sentence.

Run the binary.

Two regions in fifteen minutes. AGPL forever, commercial when you ship.

Get it on GitHub Sign up for Cloud

Not deploying today? Drop your address and we’ll send launch notes.