Skip to content
KanshiKanshi Docs

What is Kanshi?

Kanshi is an open-source, self-hosted observability platform. A lightweight Go agent collects CPU, memory, disk, and network metrics from your hosts. Optional process telemetry adds per-process CPU and resident memory. Instrumented applications send traces and correlated logs through an OpenTelemetry Collector. A central core stores telemetry in TimescaleDB, serves the real-time React dashboard, evaluates alert rules, and delivers firing and resolved transitions to webhooks.

The pieces

Section titled “The pieces”
Component What it does Runs as
agent Samples host and optional process metrics, batches them, ships them over gRPC One binary per monitored host
core Ingests metric batches, persists them to TimescaleDB, exposes a REST API One service (gRPC :50051, REST :8080)
dashboard Real-time charts per agent and fleet overview Static React app (Nginx in Docker)
OpenTelemetry Collector Batches, authenticates, and forwards application traces and logs One private container or service

The current stable Agent, Core, and Dashboard release is v1.3.1.

Design principles

Section titled “Design principles”
  • Small surface area. One agent, one core, one dashboard. No message broker, no sidecar zoo, no external SaaS dependencies.
  • Time-series native. Storage is TimescaleDB, PostgreSQL with time_bucket rollups, so range queries stay fast as history grows and you can query up to seven days of host history with plain SQL.
  • Self-hosted. Your metrics never leave your infrastructure.
  • Private by default. Process telemetry is disabled until you explicitly enable it. Shared OTLP identity attributes support navigation and are not an authorization boundary.

Next steps

Section titled “Next steps”