Silos Replication · WAL streaming, rolling out to 100+ regions

WAL streamed to the edge.
Designed for ~50–200ms propagation.

Stream PostgreSQL WAL changes to edge regions worldwide — engineered for ~50–200 ms propagation — with tunable consistency on every query. Deploy to hosted endpoints or integrate via the SDK.

Get started
✓ NATS JetStream✓ Per-query consistency✓ LSN read-your-writes
console.silos.sh · replicationNATS
WAL streamLSN 0/1A3F88 · JetStream
primary us-east-1
eu-west-1applied 0/1A3F70lag 42 ms
ap-tokyoapplied 0/1A3F61lag 88 ms
sa-east-1applied 0/1A3F2Clag 130 ms
Consistency · per query
Lag values illustrative · ~50–200 ms propagation target
~50–200 ms
WAL streaming target
100+
regions (rolling out)
NATS
JetStream WAL transport
4
tunable consistency levels
Realtime replication

Replication built for scale.

WAL streaming with tunable consistency — the same pipeline your agents drive through the API.

Try the demo Clone replication config

WAL-based streaming architecture

Native PostgreSQL WAL capture, designed to stream over NATS JetStream — an ordered, durable transport for delivering changes to edge replicas.

PRIMARY
WAL
NATS
stt_capture
>
wal_decoder
>
nats_publisher
>
edge_materializer
>
replica_writer
>

Configurable consistency for every query

Choose the right consistency level per query. Monitor replication lag across every region in realtime.

Learn more
Avg replication lagLast 30s

Per-region deployment controls

Instantly enable new regions and configure replication policies. Pin tables, set TTLs, and control which data goes where.

Enable region
US East
4ms
EU West
12ms
AP South
38ms
SA East
85ms

One primary. Many edges.
Designed to land changes in ~50–200ms.

Capabilities

Continuous replication,
zero compromise.

Single-writer WAL streaming with consistency you tune per query — and failover designed in for when a region goes dark.

Low-latency target
WAL streaming over NATS JetStream is engineered for ~50–200 ms propagation to edge replicas — a design target as the streaming path matures.
100+ regions (rolling out)
On-demand replicas are designed to materialize at the edge closest to your users, then scale to zero when idle. Region coverage is expanding.
Four consistency levels
Choose per query: eventual, read-your-writes (default), bounded-staleness, or strong for transactions.
Failover (roadmap)
Designed so reads route to the nearest healthy replica when a region is unavailable. Automatic promotion is on the roadmap.
WAL-based CDC
Built to stream changes to external systems and power reactive pipelines straight from database events.
Tunable per query
Set the consistency level per statement — mix eventual reads with strong writes in one session.
Global consistency, your choice

Read-your-writes
session consistency.

The default. Writes return an LSN token; reads pass a min-LSN, and the router waits for the replica to catch up — or routes to the primary — so a client always sees its own writes.

Read routing · session-aware
Live in the console

Watch replicas hydrate.
Read lag in realtime.

Promote replicas, tune consistency, and stream live metrics — without leaving the browser. The same console your agents drive through the API.

db-f7k2 · production
console.silos.sh
Silos
all systems healthy
Live metrics streaming · last 60s
Queries / sec
0
p99 latency
0.0 ms
Active conns
0
Storage
84.2 GB
Why Silos Replication

Latency, availability,
and consistency you tune.

Your users deserve fresh data, everywhere — without operational overhead.

Designed for fast propagation

WAL streaming is engineered to reach edge replicas in ~50–200 ms, so users get near-fresh data wherever they are. A design target as the streaming path matures.

Propagation (illustrative)

Failover by design

Designed so reads route to the nearest healthy replica when a region is unavailable. Automatic promotion is on the roadmap.

us-east-1 (primary)
eu-west-1
ap-south-1
us-west-2

Query-level consistency

Choose eventual, read-your-writes, bounded-staleness, or strong consistency on every query. No global locks.

eventualnearest replica
read-your-writesdefault
bounded-stalenesscapped by age
strongprimary read
Designed for

Built for a range of
production workloads.

The architecture targets transactional, multi-region, and event-driven use cases. Silos is pre-GA — these describe what the system is designed to do.

Transactional

Strong, primary-routed reads for financial and transactional workloads that can't tolerate stale data.

Global SaaS

One primary, on-demand edge replicas — designed to scale from a single region to many without application changes.

Event pipelines

WAL-based change data capture, built to feed reactive pipelines and analytics straight from database events.

Replication patterns

Built for every
replication pattern.

From edge caching to offline-first sync — Silos Replication is designed for a wide range of production patterns.

01

Edge API caching

Materialize read replicas at the edge to serve API responses close to your users. Replicas stay current via WAL streaming, with staleness you bound per query.

02

Multi-region SaaS

Deploy your PostgreSQL database once and replicate to the regions where your customers operate, with region-aware routing and failover designed in.

03

Realtime analytics

Stream WAL changes into analytics pipelines for realtime dashboards, CDC integrations, and event-driven architectures without polling.

04

Offline-first sync

Combine with PGLite for browser-to-edge bidirectional sync. Users work offline, changes merge automatically when connectivity returns.

FAQ

Questions, answered.

Silos Replication builds on PostgreSQL's WAL but is designed to stream changes over NATS JetStream for ordered, durable delivery to edge regions. The model adds on-demand replica materialization and consistency levels you tune per query — going beyond native logical replication. Silos is pre-GA, and the end-to-end streaming path is still being hardened.
Four levels: eventual (read from the nearest replica, may be stale), read-your-writes (the default — session-aware via LSN tokens), bounded-staleness (read data no older than a bound you set), and strong (always read from the primary). You choose per query or per connection.
Each write returns a log sequence number (LSN). Reads can pass a minimum LSN, and the router waits for the target replica to catch up to that point — or routes the read to the primary — so a client always observes its own writes. This is the default consistency level.
Silos is designed to monitor replica health and replication lag and route reads to the nearest healthy replica when a region is unavailable. Automatic primary promotion is on the roadmap rather than a guaranteed, production-proven capability today.
Install the Silos CLI, connect your database, and enable replication from the CLI or console. On-demand edge replicas are designed to materialize based on your configuration. Silos is pre-GA — expect the experience to keep evolving.
Teams running PostgreSQL that want lower read latency, broader availability, or global distribution. Target use cases include SaaS platforms, API backends, edge computing, event pipelines, and offline-first applications.
Silos speaks the PostgreSQL wire protocol and runs real PostgreSQL, so standard drivers, ORMs, and migration tools connect the way they would to any Postgres. Full external-database replication and migration tooling is part of the evolving roadmap.
WAL streaming is engineered for roughly 50–200 ms propagation; same-continent replicas should land at the lower end. These are design targets, not benchmarked guarantees. You can monitor lag per region via the dashboard or a Prometheus metrics endpoint.

Start streaming.

Connect your primary, configure regions, and stream WAL to the edge. Silos is pre-GA — start building today.