SELF-HOSTED REALTIME MESSAGING

The realtime messaging engine you deploy yourself

One Go binary powers group chats, live rooms and direct messages. Your data stays on your servers; add AI and enterprise ops when you need them.

Core capabilities

💬

Three conversation types

Groups, live rooms and DMs

Complete messaging

Reply, recall, receipts, reactions

🔀

Horizontal scaling

RabbitMQ / NATS brokers

🤖

AI agents

RAG, tools, human handover

🎛️

Admin console

RBAC, audit, ops trends

📞

A/V signaling

1-on-1, multi-party, PK

🔐

Data sovereignty

Self-hosted MySQL / PostgreSQL

🧩

Integrations

Account sync, gRPC API

One codebase, from a single node to a cluster

Clients websocket ×N Broker RabbitMQ / NATS persist_consumer ai_worker sync_worker MySQL / PostgreSQL · Redis

Real benchmarks · no inflated claims

Single-node performance: reproducible numbers, not theoretical peaks

1,000
concurrent connections
steady step with zero errors / drops
3,636
msg/s fan-out
actual group-message delivery rate
70 ms
p99 latency
p50 just 18ms
0
errors / drops
no slow-consumer closes, no throttling

Test environment (single instance)

Instance
Alibaba Cloud ECS · 4 vCPU / 8 GiB (hfc6.xlarge)
Process
One websocket binary · single-node direct delivery (no broker)
Storage
PostgreSQL + Redis (same host)
Scenario
Multi-room group chat (50 rooms) · steady step of a saturation sweep

Why these numbers hold up

We hand you the load-test tooling

We provide the load generator and report tool (loadtest / satreport) so you can rerun the exact parameters, scenarios and result JSON in your own environment — reproducible on your side, not just our word.

Modelled on real load

Traffic is shaped like real users typing (within the server’s 4 msg/s per-user limit), not one connection hammering to inflate throughput.

Steady values, not the edge of collapse

We publish the zero-error, zero-drop step; past the knee the server degrades gracefully (2,000 conns still sustain 7,000+ msg/s fan-out, p99 < 100ms) instead of cascading.

A single node is just the floor

RabbitMQ / NATS scale out horizontally; capacity grows near-linearly with websocket instances — multiply the single-node figure by your instance count.

Measured on Alibaba Cloud cn-hangzhou, 2026-07. Different instance types, message sizes and room topologies will vary — rerun the benchmark for your own workload shape.

Four editions, pick what you need

Basic

Single-instance IM core, self-hosted starter

Pro

Cluster-scalable with push / moderation / callbacks — production-ready

Customer Service

Pro + the full AI customer-service suite

See the full comparison and pricing →

Security & private deployment

Data sovereignty · self-hosted, never leaves

Auth & audit · JWT / RBAC / audit log

Encrypted transport · WSS / TLS

Intranet / offline capable

Security & private deployment →

Ready to own your messaging infrastructure?

Contact us Read the docs →