In active development · MIT

Replace Sentry with one container you own.

OpenFault is an error tracker your existing SDKs already speak. Change the DSN and you’re done. It idles under 100 MB of RAM, it’s MCP-first so your agents can triage, and it’s open source.

docker run -d -p 8080:8080 -v openfault:/data ghcr.io/openfault/openfault

Get started Star on GitHub

errors.example.com/acme/web/issues Preview

is:unresolved env:production Last seenUsersEvents

Issue	Events	Users	Seen
TypeError Cannot read properties of undefined (reading 'id')Regressed app/checkout/[id]/page.tsx in loadOrder	1,204	212	2m
ActiveRecord::RecordNotFound Couldn't find Invoice with 'id'=<int>New app/controllers/invoices_controller.rb in show	86	41	14m
Net::ReadTimeout Net::ReadTimeout with #<TCPSocket:(closed)> app/jobs/sync_stripe_job.rb in perform	532	3	1h
ZeroDivisionError division by zero billing/usage.py in proration	19	7	3h
Error Hydration failed because the server rendered HTML didn’t match components/PriceTag.tsx in PriceTag	3,890	1,018	5h

01 · Drop-in

Keep your SDKs. Change one line.

OpenFault speaks the Sentry envelope protocol. The official Sentry SDKs post to it as they are. No fork, no plugin, no wrapper.

Rails
Next.js
Node
Python

Minified stack traces? sentry-cli sourcemaps upload works against it, same flags, same debug IDs.

Not doing Tracing, session replay, profiling. On purpose. OpenFault tracks errors. That is the whole job.

sentry.client.config.ts

Sentry.init({
-  dsn: "https://k3y@o1234.ingest.sentry.io/56",
+  dsn: "https://k3y@errors.example.com/1",
   tracesSampleRate: 0,
});

02 · Grouping

Grouping that doesn’t split bugs.

Issues are keyed on the shape of the stack: module, file, function. Never line numbers. Add a log line above the crash and the next event still lands in the same issue.

A second key on exception type and message catches refactors that move the stack. Want your own rule? Send a fingerprint. {{ default }} keeps our key in the mix.

Two events, one issue

evt a1f3order.rb:42Order#total

evt 9c07order.rb:57Order#total

keyorders | order.rb | total#4812

Custom fingerprint

Sentry.captureException(err, {
  fingerprint: ["{{ default }}", tenant.region],
});

03 · MCP

Built for agents. Gated by people.

MCP is built in. Every tool calls the same handler the UI uses, so an agent sees what you see and nothing more.

An agent can list, search and triage. It can’t read the stack or start a fix until a person clicks Approve. Captured data is untrusted: every message, breadcrumb and tag comes back fenced, with a note telling the model not to follow it.

list_new_issues
search_issues
get_issue
resolve_issue
mute_issue
link_fix

Claude Code

claude mcp add -t http openfault https://errors.example.com/mcp -H "Authorization: Bearer $OPENFAULT_TOKEN"

get_issue 4812

TypeError in app/checkout/[id] · 212 users
first seen 1.42.0 · last seen 2m ago

Fix not approved. Ask a person to approve it in OpenFault.

04 · One container

One container. One volume. That’s the stack.

A Rust binary and SQLite. No Redis, no Kafka, no Postgres, no ClickHouse. It runs next to your app on the small VPS you already pay for.

Image: ~12 MB
RAM at 4k errors/day: < 100 MB target
Default retention: 90 days
Backup: /data

05 · Teams

Teams, email, iOS.

It’s for the whole team, not just whoever set it up.

Invites and roles: Owner, admin, member. Invite by email or send a link. Closed signup by default.
Alerts: Email, Slack, Discord, or any webhook. On new issues, regressions and spikes, per person, per project.
iOS Coming: A native app with home and lock screen widgets, and push with Resolve and Mute right on the notification. Pair it by scanning a QR code.

06 · Compare

How it stacks up.

OpenFault numbers are design targets until the first release. A dash means we haven’t checked.

	OpenFault	Sentry self-hosted	Bugsink
Containers to run	1	10+	1
RAM	Under 100 MB (target)	16 GB + 16 GB swap minimum, 32 GB recommended	—
Runtime	Rust + SQLite	Python, Kafka, ClickHouse, Postgres, Redis	Python
Sentry SDK compatible	Yes	Yes	Yes
Source maps	Yes, via sentry-cli	Yes	Yes, via sentry-cli
MCP server	Built in, approval-gated	Separate server	—
iOS app	Coming, with widgets and push	—	—
License	MIT	FSL-1.1-Apache-2.0	PolyForm Shield 1.0.0

07 · Install

Running in a minute.

One command, no config. The secret key is generated into /data on first boot.

Start the container.

docker run -d -p 8080:8080 -v openfault:/data ghcr.io/openfault/openfault
Open http://localhost:8080. The setup wizard creates your admin account and first project.
Copy the project’s DSN into your Sentry SDK config. Deploy. Watch the first error come in.

08 · FAQ

Questions.

Is it really drop-in?

For errors, yes. OpenFault accepts the same envelope endpoint the Sentry SDKs already post to. Point the DSN at your server and events land. Items we don’t handle, like transactions, are dropped, not rejected, so your app never notices.

What about performance tracing?

Not planned. No tracing, no session replay, no profiling. If you need those, keep a tool for them and send errors here. Doing one job is how this stays one small container.

Can I migrate from Sentry?

Change the DSN in each app and deploy. New errors flow into OpenFault from then on. We don’t import history, so keep Sentry around read-only until you stop needing the old issues.

How is data retained?

Events are kept for 90 days by default, set per project. Within an issue we keep the first 25 events of each hour, then 1 in 20, and the counters always count. A noisy bug can’t fill your disk. Everything lives in /data.

Is it production ready?

Not yet. OpenFault is in active development and nothing here is ready for production. Follow the repo on GitHub to see when the first release ships.