Why it matters
Modern observability platforms charge per host, per seat, and per event, layers that compound fast on any real workload. Maple inverts that model: pay per GB ingested, no per-host or per-seat fees, source available under FSL-1.1 on GitHub.
The big picture
Maple is an end-to-end observability backend for distributed systems. It collects traces, logs, and metrics over standard OTLP, stores and queries them at ClickHouse speed, and surfaces them through a unified dashboard covering service maps, trace waterfalls, structured log streams, and Kubernetes pod heatmaps.
How it works
Send OTLP from any instrumented service to Maple's ingest gateway. The gateway applies key auth and org enrichment, then forwards to the collector. From there, ClickHouse handles storage and query. The frontend, built on TanStack Router, renders live service maps, trace trees, and metric charts. Dashboards are persisted in SQLite (local) or Turso (cloud). Per the README, the local binary bundles OTLP ingest, embedded ClickHouse, and the dashboard in a single maple start command installable via Homebrew.
Zoom in
The MCP server is a first-class feature, not an afterthought. Any compatible AI agent can call list_services, find_errors, error_detail, and propose_fix over the open MCP protocol. The site demo shows an agent diagnosing a Stripe idempotency collision and opening a PR in 5 tool calls across 4.2 seconds.
Yes, but
FSL-1.1 is not OSI-approved open source. It restricts certain competitive uses until the code converts to Apache 2.0 (the timeline is set by the license terms, not a public date). Teams with strict open-source procurement policies should review the FSL terms before adopting.
The bottom line
Maple is the observability stack to reach for when you want OTel-native ingest, honest GB-based pricing, and an AI agent that can read your telemetry and act on it, all from a codebase you can fork and self-host.