About PrismaLens

Why We Built This

At 2:47am on a Tuesday, our payment API started timing out. Revenue dropped to zero. The on-call engineer spent 90 minutes jumping between Grafana dashboards, grepping logs, and correlating deployment times before finding the cause: a database migration that added an unindexed column.

That engineer had all the right tools — Prometheus, Grafana, distributed tracing, structured logs. The problem wasn't observability. It was correlation. Connecting the dots across systems, timelines, and signal types is cognitive work that takes time and drains focus.

The PrismaLens CLI automates that correlation. Point it at a repo, pipe in a firing alert (or describe the problem with --query), and it rents a harness — a coding agent it drives locally — to gather evidence read-only, then hands back a report: hypotheses ranked most to least plausible, each backed by the evidence that supports or contradicts it — no numeric confidence scores, just an ordered trail you can check.

We built PrismaLens because we believe incident response should be about fixing problems, not finding them. Your observability tools already captured the evidence. Let AI do the detective work.

Open Source by Design

PrismaLens is open source (Apache-2.0) because incident response tools should be transparent. You should know exactly how investigations run, what data gets queried, and how conclusions are reached. Run it with your own model provider key — no PrismaLens account, no subscription, and every run's events and report stay on your machine under ~/.prismalens/runs/.

Built for SREs, by SREs

We're not building another observability platform. The CLI is a terminal-first tool that reads your repo and investigates read-only — it doesn't ask you to migrate off Prometheus, Grafana, Datadog, or New Relic. A self-hosted server that connects to those tools directly is on the roadmap, but not released yet.