Self-hosted user behavior analytics, session replay, and bug reporting for small closed-loop apps. One Go binary, one SQLite file, one npm package.

spyglass is the telemetry stack for internal tools, the apps with 20–200 identified daily users that need events, session replay, error tracking, and bug reports, but none of the billion-event machinery (ClickHouse, Kafka, Kubernetes) that PostHog, Highlight, and OpenReplay assume.

It is one cross-compiled Go binary writing to one SQLite file plus gzipped replay blobs on disk, paired with a ~5KB npm SDK. Zero external services, no database server, no phone-home. Everything stays on the operator's machine.

These docs are the source of truth and live alongside the code. They are authored as Fumadocs MDX and rendered by a separate website repo.

The structural advantage

In a closed internal system, every session is recorded continuously with identified users. "What happened when the bug occurred" stops being a capture problem and becomes a query over data already on disk, so bug reporting is a view layer, nearly free.

Start here

Getting started

Run the collector, wire the SDK, see your first events, in about five minutes.

SDK reference

init, capture, setUser, report, and every config option.

Configuration

The spyglass.config.json file in full: apps, keys, retention, auth.

Dashboard

A tour of every view, including the incident slice.

Deployment

Docker, the static binary, auth, retention, and backup.

API reference

Every collector HTTP endpoint, with parameters and responses.

How it fits together

@spyglass/sdk  ──batched JSON / gzipped replay──▶  spyglassd (one Go binary)
(in your app)                                       ├─ SQLite (events + sessions)
                                                    ├─ disk (gzipped replay chunks)
                                                    └─ embedded dashboard at /

The SDK (~5KB gzipped) records events, replay, network, and errors and ships them to the collector. The collector stores everything in one SQLite file plus replay blobs on disk, and serves the dashboard from the same binary. That's the whole system. See Getting started to stand it up.

Documentation