Multi-region key value infrastructure built on open-source Valkey. The same active-active replication, conflict resolution, and regional resilience that proprietary platforms charge six figures for — without the licensing costs, vendor lock-in, or opaque pricing.
You need active-active replication across regions. The incumbent enterprise solution prices it per shard, per region, with annual contracts. For most teams, the licensing alone exceeds the infrastructure cost.
Indicative pricing based on publicly available enterprise vendor rate cards and typical cloud compute costs (3 regions, 2 replicas per region, 50 GB working set). Actual costs vary by provider, region, and configuration. These figures are estimates — not a binding quote.
AtlasKV puts a lightweight replication proxy in front of standard Valkey instances. All multi-master semantics — conflict resolution, distributed counters, cross-site replication — live in the proxy. Valkey stays unmodified.
Upgrade Valkey independently of AtlasKV. The replication stream runs on Kafka or Redpanda that your team already operates. If you stop using AtlasKV, your Valkey instances keep running with your data intact.
Applications connect using standard RESP protocol client libraries. No proprietary SDK. No code changes beyond pointing at the proxy endpoint.
HLC timestamping for deterministic conflict resolution. Last-write-wins for strings and hash fields. Conflict-free distributed counters for INCR/DECR workloads. Tombstone-based deletes with cross-site propagation. Owner-driven TTL expiry to prevent resurrection races.
SITE A SITE B
┌─────────────────────────┐ ┌─────────────────────────┐
│ Clients (RESP) │ │ Clients (RESP) │
│ │ │ │ │ │
│ ▼ │ │ ▼ │
│ ┌────────┐ ┌───────┐ │ WAN │ ┌────────┐ ┌───────┐ │
│ │ Proxy │──▶│ OpLog │─┼─Replicate──┼─▶│Applier │──▶│Valkey │ │
│ │ │ │ │ │ │ │ │ │ │ │
│ └───┬────┘ └───────┘ │ │ └────────┘ └───▲───┘ │
│ │ │ │ │ │
│ ▼ │ │ ┌────────┐ │ │
│ ┌────────┐ │ │ │ Proxy │───────┘ │
│ │ Valkey │◀────────────┼─Applier◀───┼──│ OpLog │ │
│ └────────┘ │ │ └────────┘ │
└─────────────────────────┘ └─────────────────────────┘
The capabilities that matter for multi-region deployments — built into the proxy, not sold as add-on tiers.
Every region accepts reads and writes at local latency. No cross-region round trips. No primary node bottleneck.
Hybrid Logical Clocks order concurrent writes consistently across all sites. No manual conflict handling. Predictable outcomes.
INCR/DECR operations that never lose updates across sites. Per-site sub-counters converge automatically. Built for metering, rate limiting, and soft quota tracking.
When a region goes down, the others keep serving. When it comes back, data converges from the replication log. No manual failover runbook.
RESP protocol compatible. Use your existing client libraries, monitoring tools, and operational knowledge. No proprietary SDK.
Replication lag, conflict events, clock skew, tombstone backlog, proxy latency. Everything you need to operate confidently.
Not every command can be replicated across sites with sane semantics. We document exactly what works, what’s local-only, and what’s coming — rather than claiming full compatibility and leaving you to discover the gaps.
| Command Family | Replication | Semantics |
|---|---|---|
| GET / SET / GETSET / strings | Full | Last-write-wins (HLC-ordered) |
| INCR / INCRBY / DECR / INCRBYFLOAT | Full | Conflict-free distributed counters |
| HSET / HGET / HDEL | Full | Last-write-wins per field |
| DEL / UNLINK | Full | Tombstone propagation |
| EXPIRE / TTL / PERSIST | Full | Owner-driven expiry |
| SADD / SREM / sets | Roadmap | Observed-remove sets (phase 2) |
| MULTI / EXEC | Local | Per-site atomicity; replicated as individual ops |
| Lists (LPUSH / RPUSH) | Local | Local-only; no sane cross-site merge |
| Pub/Sub | Local | Per-site; cross-site bridging on roadmap |
Any workload where the cost of enterprise active-active licensing doesn’t match the value of the data being replicated.
Serve users from multiple regions with a unified data layer and local-latency access. No cross-region reads on the hot path.
User sessions available across datacenters and cloud providers. Failover without session loss. No sticky routing required.
Active-active cache replication that survives regional outages without cold-start penalties or origin thundering herd.
Distributed counters for usage tracking, rate limiting, and soft quota enforcement across regions. Per-site budget allocation for overshoot control.
Subscriber state, session data, and operational counters for distributed network services. Regional resilience for carrier-grade workloads.
Same replication layer across cloud providers, on-premises, and hybrid environments. No cloud-specific lock-in.
AtlasKV is free. Installation and architecture consultancy by TENSORY³ is mandatory — we deploy it correctly into your environment, configured for your topology, tested under your workload. After go-live, support is optional.
Replication proxy, conflict resolution, distributed counters, and observability. Free. No licensing fees. No per-shard charges.
Architecture review, deployment into your infrastructure, replication topology design, workload testing, and production sign-off. One-time engagement.
Monitoring, incident response, upgrade assistance, and operational guidance. Monthly retainer after go-live. Scale up or cancel as needed.
We run the platform end to end. Installation, operations, monitoring, upgrades. You build applications. We handle the infrastructure.
AtlasKV is an AP system. Concurrent writes to the same key at different sites are resolved deterministically — but one write wins and the other is dropped. This is the correct behaviour for cache, session, and configuration workloads.
For workloads requiring strong consistency, globally atomic transactions, or hard quota enforcement with zero overshoot, AtlasKV is not the right tool. We say so upfront rather than burying it in footnotes.
Lists, Lua scripts, and blocking commands are local-only. MULTI/EXEC gives per-site atomicity, not cross-site. Sets get full CRDT replication in phase 2.
We publish the exact command compatibility matrix because discovering replication gaps in production is worse than knowing them at evaluation time.
Tell us about your deployment — regions, dataset size, workload type — and we’ll scope the installation and send you a cost comparison against your current setup. No sales deck. Just numbers.
Or talk to us directly. No procurement cycle required.