Teams either send production error data to a hosted service at volume-based pricing, or run a heavy self-hosted stack, or monitor nothing. We built a Sentry-compatible tracker that ships as a single binary and adds LLM root-cause analysis, with pluggable model backends including fully local inference. Migration is a DSN change.
Error monitoring presents development teams with an awkward choice. Hosted services work well but send production error data — including stack traces, request context and whatever ends up in them — to a third party, at a price that scales with event volume. Self-hosting the same tools means running a substantial infrastructure stack for what is, functionally, a queue and a database. Many teams resolve this by monitoring nothing, which is the worst option available. Compounding it, monitoring tells you an error occurred but not why: triage still consumes senior engineering time on the same recurring question of what actually broke. A viable alternative had to be genuinely trivial to run, and had to do more than collect.
The service speaks the Sentry protocol, so migration is a DSN change rather than a code change. Existing Sentry SDKs in any language keep working untouched — the single largest barrier to switching error trackers simply does not apply.
Ships as one binary with no Docker Compose stack, no message broker and no separate worker fleet. A team can run production-grade error tracking on hardware they already have, which is the difference between self-hosting in principle and self-hosting in practice.
Incoming errors are analysed by an LLM to propose a likely root cause alongside the stack trace, moving triage past 'this broke' toward 'this broke because'. The analysis targets the recurring, low-novelty errors that consume disproportionate senior attention.
Works with Claude, OpenAI, Ollama and YandexGPT, so teams choose between hosted model quality and fully local inference. For organisations that self-host specifically to keep data in-house, local models keep that guarantee intact end to end.
Teams get a drop-in replacement for a hosted error tracker that keeps production data on their own infrastructure, deploys as a single binary, and adds root-cause analysis their previous tool did not provide. Because ingestion is Sentry-compatible, the switch costs a configuration change rather than a migration project. The project is open source and under active development.
Tell us what the process looks like today and we will tell you what can be automated — and what should not be.
LET'S TALK
