| llms_description | Security and data (server 2.11.0): what leaves the machine — server: update check GET api.github.com/repos/buggregator/server/releases/latest at start and every 12h (User-Agent buggregator/<version>; UPDATE_CHECK_ENABLED=false turns it off), Sentry JS source context: server GETs http(s) URLs from JS stack-frame filenames (3 s, 2 MiB, TLS not verified, any host in 2.10.0 and older; since server 2.11.0 loopback/private only by default, sentry.source_fetch / SENTRY_SOURCE_FETCH / --sentry-source-fetch = private|any|off); browser: UI fonts ship with the UI since 2.11.0 (2.10.0 and older load Google Fonts on every open); images and styles inside captured HTML emails and HTML dumps load from their hosts when opened (scripts do not run), no switch; only when you turn them on: SSO provider, webhooks, OpenAPI import, HTTP proxy upstreams, Xdebug IDE. Since server 2.11.0: «Check now» calls releases/latest on click (≤1/min); «Update to X» downloads the release binary, checksums.txt, checksums.txt.sig from GitHub release assets (or the UPDATE_RELEASE_URL mirror), UPDATE_SELF=off. No analytics or usage reporting. Listening: binary on all interfaces (:8000, :1025, :9912, :9913, HTTP proxy :8080 = open forward proxy with HTTPS MITM, no auth); opt-in: MCP over HTTP :8001 all interfaces, no auth without MCP_AUTH_TOKEN (MCP_ADDR to bind), METRICS_ADDR; bind with -http-addr/-smtp-addr/-vardumper-addr/-monolog-addr and PROXY_ADDR=127.0.0.1:8080; Docker: publish as 127.0.0.1:port:port. SSO protects UI and UI APIs only: ingestion, TCP ports, HTTP proxy, /mock/, /api/version, /api/settings, attachment links, MCP HTTP, /metrics stay open. Data: in memory by default (lost on restart), DATABASE_DSN file, attachments STORAGE_MODE; no automatic cleanup by default; since server 2.11.0 retention.max_age / RETENTION_MAX_AGE / --retention-max-age (30m, 24h, 7d) and retention.max_events / RETENTION_MAX_EVENTS / --retention-max-events, default keep all, every minute, pinned kept and not counted, attachments go too; max_age also prunes Sentry logs/metrics (received) and traces/spans (start time), max_events counts event cards only; deleting an event also deletes its Sentry error details; manual clear of events keeps Sentry logs/metrics/traces (use DELETE /api/sentry/all). Versions: semver, minor may change behaviour, pin ghcr.io/buggregator/server:2.11.0 (no v), tags latest, 2, 2.11, 2.11.0; fixes only for the latest release line. |
|---|
Buggregator is a debugging tool for your own machine or a trusted dev network. It has no accounts, sends no analytics and reports no usage. This page lists, without gaps, what it does send, who can reach it, and where your events are kept. Everything here is true for server 2.11.0.
| What | Who sends | Where to | When | How to turn it off |
|---|---|---|---|---|
| New-version check | the server | https://api.github.com/repos/buggregator/server/releases/latest |
at start, then every 12 hours | UPDATE_CHECK_ENABLED=false |
| Source of JavaScript errors | the server | the URL of each file in a JavaScript stack trace | each Sentry error from a browser or JS app | SENTRY_SOURCE_FETCH=off (since 2.11.0) |
| Images and styles inside captured HTML | your browser | the hosts named in that HTML (e.g. a tracking pixel) | when you open an HTML email or an HTML HTTP dump | no switch |
New-version check. One GET request. Besides your IP address, it carries only the server version
(User-Agent: buggregator/2.11.0). Nothing about your events or projects is sent. With
UPDATE_CHECK_ENABLED=false the server makes no such call at all. Details:
Server → New-Version Check.
UI fonts. Since server 2.11.0 the fonts (DM Sans, JetBrains Mono) are part of the UI, and the browser
loads nothing from Google Fonts. Server 2.10.0 and older load them from fonts.googleapis.com and
fonts.gstatic.com, so Google sees the IP address of the browser that opens the UI. No event data goes there.
Source of JavaScript errors. Browser SDKs send stack frames with a file URL but no code, e.g.
http://localhost:5173/src/App.vue. To show the code, the server downloads that file: up to 2 MiB, 3-second
timeout, TLS certificates not checked. In server 2.10.0 and older it does this for any http:// or https:// URL
in a frame, not only for local ones. So whoever can send a Sentry event to your Buggregator can make it request a
URL from inside your network. Keep the server where only your apps can reach it (see below).
Since server 2.11.0, the server fetches only from loopback and private addresses by default
(127.0.0.0/8, ::1, 10.0.0.0/8, 172.16.0.0/12, 192.168.0.0/16, fc00::/7): your dev server, a Docker
service. One setting changes it: sentry.source_fetch in buggregator.yaml, SENTRY_SOURCE_FETCH or
--sentry-source-fetch, with private (default), any (every host, as in 2.10.0 and older) or off (never fetch).
The check is made on the resolved IP address, cloud metadata addresses are refused, and redirects go only to the
same host. Details: Server → Sentry Source Fetch.
Images and styles inside captured HTML. An HTML email or an HTML HTTP dump is shown as it is. Scripts in it do not run, but its images and stylesheets load from wherever they point. So when you open a captured email, your browser requests its images, tracking pixels included, and the sender may see that the email was opened.
| What | Who sends | Where to | When | How to turn it off |
|---|---|---|---|---|
| Check now (since server 2.11.0) | the server | the same GitHub address as the new-version check | when you click it, at most once a minute | UPDATE_CHECK_ENABLED=false |
| Update to <version> (since server 2.11.0) | the server | the release files on GitHub: the program for your system, checksums.txt, checksums.txt.sig |
when you click it | UPDATE_SELF=off |
UPDATE_RELEASE_URL (update.release_url) points both at a mirror of GitHub's releases/latest instead. The
download is still checked against the signing key built into Buggregator. See Install → Update.
| Feature | Where it connects |
|---|---|
| SSO | your login provider (GitHub, Google, Auth0, Keycloak, GitLab, OIDC) |
| Webhooks | the URLs you configure |
| Mock server OpenAPI import from a URL | the URL you enter |
| HTTP proxy | the hosts your app sends through it |
| Xdebug proxy | your IDE's address |
Nothing else makes outgoing connections.
Every port listens on all network interfaces by default. Anyone on your network who can reach your machine can open the UI, read your events and send new ones.
| Port | What | Default |
|---|---|---|
8000 |
web UI, API, HTTP events | :8000, all interfaces |
1025 |
SMTP | :1025, all interfaces |
9912 |
VarDumper | :9912, all interfaces |
9913 |
Monolog | :9913, all interfaces |
8080 |
HTTP proxy | :8080, all interfaces |
9003 |
Xdebug proxy (off by default) | 127.0.0.1:9003 |
8001 |
MCP over HTTP (only with MCP_ENABLED=true and MCP_TRANSPORT=http) |
:8001, all interfaces |
METRICS_ADDR |
Prometheus metrics on their own port (only when you set it) | the address you set |
MCP over HTTP asks for no token unless you set MCP_AUTH_TOKEN. Bind it with MCP_ADDR=127.0.0.1:8001. The
default MCP transport is a local socket file, which opens no port.
The HTTP proxy deserves a note. It forwards requests to any host, also HTTPS, and asks for no password. On all interfaces, it is an open proxy for everyone on your network.
To keep everything on your machine:
PROXY_ADDR=127.0.0.1:8080 ./buggregator \
-http-addr 127.0.0.1:8000 \
-smtp-addr 127.0.0.1:1025 \
-vardumper-addr 127.0.0.1:9912 \
-monolog-addr 127.0.0.1:9913The same with environment variables: HTTP_ADDR, SMTP_ADDR, VAR_DUMPER_ADDR, MONOLOG_ADDR, PROXY_ADDR.
Inside the container the server listens on all interfaces; that is fine. What matters is how you publish the
ports. Always put 127.0.0.1: in front:
ports:
- "127.0.0.1:8000:8000" # only this machine
# - "8000:8000" # every machine that can reach yoursContainers on the same compose network reach Buggregator by its service name. They need no published ports at all.
Publish only what you open from your machine: usually just 8000.
SSO puts a login in front of the UI and the APIs the UI uses. These stay open, also with SSO on:
- event ingestion (Sentry, Ray, Inspector, profiler, SMS, HTTP dumps) and the SMTP, VarDumper and Monolog ports;
- the HTTP proxy;
- the mock server under
/mock/; /api/version,/api/settingsand attachment links;- MCP over HTTP on
:8001: it has its own token,MCP_AUTH_TOKEN; - Prometheus
GET /metrics, when metrics are on: on port8000, or onMETRICS_ADDRif set.
So SSO keeps strangers from reading your events, but not from sending them. On a shared server, also limit who can reach the ports, e.g. with a firewall or a private network.
Scripts call the protected API with the token the UI gets after login (30 days). See SSO → Calling the API from Scripts.
| What | Default | To keep it on disk |
|---|---|---|
| Events | in memory: gone when the server stops | DATABASE_DSN=/data/buggregator.db (SQLite file) |
| Email attachments, HTTP dump files | in memory | STORAGE_MODE=filesystem, STORAGE_PATH=… |
Details: Database.
How long. By default the server deletes nothing by itself. Events stay until you clear them in the UI, call
DELETE /api/events, or restart a server that keeps them in memory. With a database file, the file grows until you
clear it.
Since server 2.11.0, you can let it delete old events by itself:
RETENTION_MAX_AGE=7d RETENTION_MAX_EVENTS=5000 ./buggregatorRETENTION_MAX_AGE(retention.max_age,--retention-max-age): delete events older than this, e.g.30m,24h,7d. Empty or0: keep.RETENTION_MAX_EVENTS(retention.max_events,--retention-max-events): keep at most this many events, the newest.0: no cap.
Nothing is deleted unless you set one of them. The cleanup runs at start and then every minute. Attachments go with their events. Pinned events are never deleted and do not count towards the cap.
What goes with them:
- Deleting an event (in the UI, with
DELETE /api/events, or by retention) also deletes its Sentry error details. RETENTION_MAX_AGEalso removes Sentry logs and metrics by the time they arrived, and traces and spans by their start time. The error details of a pinned event stay.RETENTION_MAX_EVENTScounts event cards only. It does not touch Sentry logs, metrics or traces.- Clearing events by hand does not remove Sentry logs, metrics and traces. Remove those with Clear on the
Sentry page or
DELETE /api/sentry/all(see Sentry → Clearing data).
In 2.10.0 and older, deleting an event left its Sentry error details in the database.
Releases are numbered major.minor.patch (e.g. 2.11.0). A minor release may still change behaviour, so read
the release notes before you upgrade.
Pin the version and upgrade on purpose:
image: ghcr.io/buggregator/server:2.11.0| Tag | What you get |
|---|---|
2.11.0 |
exactly this release, never changes |
2.11 |
the newest 2.11.x patch |
2 |
the newest 2.x release |
latest |
the newest release, whatever it is |
The image tags have no v: 2.11.0, not v2.11.0. Release pages and binaries use v2.11.0.
Security fixes go into the newest release only. To get a fix, upgrade.
See SECURITY.md in the server repository.