Self-hosted error tracking and privacy-friendly analytics in one small binary.
Status: v0.2 "analytics", an early release (Docker image only). Error tracking works end to end in 26 languages: the official Sentry SDKs send events, which are scrubbed of personal data, grouped into issues and stored; issue and event pages with search, live updates, stack traces and triage; iOS and macOS crashes symbolicated from dSYMs uploaded with
sentry-cli; alerts for new issues, regressions, spikes and a release's first crash to ~40 notification channels (Slack, Discord, Teams, Telegram, ntfy, email, webhooks …); release health with crash-free sessions and users; and a REST API, OpenAPI document and MCP tools for all of it. Web analytics: a script under 2 KB, no cookies and no stored IP addresses, a Web tab with visitors, pageviews, pages, referrers, UTM, browsers and devices, filters and live current visitors, and it respects Global Privacy Control. App analytics with the KestrelKit Swift package: an App tab with daily, weekly and monthly active users, sessions, signal types, version adoption, OS versions, devices and locales, a signal explorer for payloads, and each release's adoption next to its crash-free rate. Installers and Unraid/Proxmox templates come by v1.0. The translations are unreviewed drafts. Sending errors to Kestrel shows how to point each SDK at it, upload dSYMs and set up alerts; Web analytics shows how to add the snippet and what is kept; App analytics does the same for KestrelKit. The design is indocs/design.md; to build and run it, seedocs/development.md.
Keen Kestrel answers two questions about the apps and sites you run:
- What broke? Errors and crashes from the official Sentry SDKs (iOS/Swift, JavaScript, Rust, Python,
Go and more) work unchanged. Issues are grouped, alerts go to the same ~40 channels (and the Hedgerow app) as
Vigilant Vole, and iOS crashes are symbolicated from the
dSYMs that
sentry-cliuploads. - How is it used? App signals and website pageviews without cookies or stored IP addresses, with dashboards for users, versions, devices, pages and referrers.
One Rust binary and SQLite: no Kafka, ClickHouse, Redis or Postgres. It shares its design language and foundations with Vigilant Vole and runs on Docker, Unraid, Proxmox or plain Linux.
Kestrel runs as one container with its data in a single folder:
mkdir keen-kestrel && cd keen-kestrel
curl -fsSLO https://raw.githubusercontent.com/Ferdinand99/Keen-Kestrel/main/docker-compose.yml
docker compose up -dThen open http://<host>:3002/setup to create the first admin, create a project and copy its DSN. Until the
first admin exists, /setup asks for a setup token: set KK_SETUP_TOKEN, or read the one Kestrel makes up and
logs at the first start (docker compose logs keen-kestrel, a line with /setup?token=…). Scripted installs
that don't want the token set KK_SETUP_OPEN=1.
Set the address your apps use under Settings → General (public URL), or as KK_PUBLIC_URL in
docker-compose.yml, which is stored there at every start: it goes into the DSNs and alert links. Put Kestrel
behind HTTPS with KK_TRUST_PROXY=1 when it's reachable from the internet. Images
are on GHCR for amd64 and arm64: ghcr.io/ferdinand99/keen-kestrel:<version> or :latest.
Back up the data folder as a whole. Notification channel secrets are sealed with the instance key in it
(session.key, since 0.6); a copy of the database without that file loses them, and they must be entered again.
Upgrading to 0.6 seals existing channels at the first start, after a plaintext copy in backups/.
The release after 0.7.0 (vole-kit 0.6) seals the two-factor secrets of your users the same way. The first start takes
a copy first (backups/, with a .plaintext-secrets marker next to it; it is deleted automatically once the database holds
no plaintext secret, or after 7 days). 0.7.0 can't open a database the new release has started on: restore that copy to
go back. Creating write or never-expiring API tokens, adding or promoting admins, an admin's password or two-factor
reset and deleting a passkey now ask for your password.
The release that adds the Hedgerow app channel renames Kestrel's internal slug from kestrel to keen-kestrel (the push
relay knows the products by their full names). It shows up as the MCP server name, the API reference download names
and the source or tag Keep, ClickSend and Opsgenie receive: update a rule that matched kestrel.
The Hedgerow app on your iPhone can show Kestrel's alerts as push notifications, with no account and no other service to run.
- Under Settings → Notifications, add a channel of the kind Hedgerow app. It has no settings. Choose it in the alert rules of your projects like any other channel. Only one Hedgerow app channel can be active at a time.
- In the Hedgerow app, open the connection to your server and turn on Varsler (notifications). The app registers itself with your server, and nothing is pushed to it before that.
Push is off until an admin adds the channel: with no active Hedgerow app channel the server refuses to register devices. A read-only API token is enough for the app. Send test on the channel shows an error until at least one device has registered. Alerts wait in the alert queue like for every channel, so a failed push is retried after 1, 5 and 15 minutes; a push that reached at least one device is never repeated.
What leaves your server is only a random reference. Kestrel keeps the text of each alert (“New issue in Shop: …”) for
24 hours and sends the Hedgerow push relay (pigeon.hedgerow.tools, or the one in KIT_PUSH_RELAY_URL / the
push_relay_url setting if you run your own) a push that carries that reference and the app's device handles,
nothing else. The relay and Apple therefore never see project names, errors or messages. When the push arrives, the
app fetches the text from your own server (GET /api/v1/push/events/{ref}) with its token and shows it. The server
registers itself with the relay the first time the channel fires, so it needs outbound HTTPS to the relay. The upgrade
to the release that adds the channel creates three empty tables (push_devices, push_events,
push_registrations) after the usual copy in backups/; the release before it refuses a database this one has
started on, so restore that copy to go back.
| Errors | Analytics |
|---|---|
| Sentry-compatible ingest (envelopes, sessions, feedback, attachments) | App signals via the KestrelKit Swift package |
| Issue grouping, search, resolve/ignore/merge, regressions | Web pageviews via a script under 2 KB |
iOS/macOS symbolication from dSYMs (sentry-cli compatible) |
Daily, weekly and monthly users, versions, devices, locales |
| Release health: crash-free sessions and users | Pages, referrers, UTM, browsers |
| Alerts: new issue, regression, spike, first crash in a release | No cookies, no stored IPs, daily-rotating salts |
| REST API, OpenAPI, MCP tools for coding agents | Same projects, keys and releases as errors |
Not in v1: performance tracing, session replay, JavaScript source maps (planned for v1.1), funnels and
cohorts. See docs/design.md for the full scope and the phases.
The October 2026 security review (what was found, fixed and left open) is in docs/security-review-2026-10.md.
Hedgerow is built in spare time and is free to use. If you find it useful, you can support it through GitHub Sponsors or Ko-fi. Thank you.
GNU AGPL-3.0-only, except the web analytics snippet
src/analytics/k.js, which is MIT so it can go on any website. The Swift SDK,
KestrelKit, is MIT too.
The licence texts of the Rust crates and web files inside the binary are in THIRD-PARTY-NOTICES. The Docker image carries the file as /usr/share/doc/keen-kestrel/THIRD-PARTY-NOTICES (next to LICENSE). The file is made by scripts/notices.sh (cargo-about, pinned) and CI fails when it is out of date or when a dependency has a licence that is not on the allowlist in about.toml. After changing a dependency, run scripts/notices.sh and commit the result.