# What PrismaLens is

> Open-source AI incident investigation on your machine, run by the coding agent you already use, with the evidence kept.

PrismaLens is an open-source app that runs on your machine. Start it with `pl up`, or open the desktop app (preview). When something breaks, it opens an incident and hands it to a coding agent you already use, such as OpenCode, Claude Code or Codex. The agent reads the service’s code, works out what went wrong and writes a report. PrismaLens keeps the whole run: what the agent did, what it found and what it ruled out.

There is no PrismaLens account, no subscription and no model bill from us. PrismaLens never calls a model. Your agent does, with its own sign-in.

PrismaLens is alpha and changes often, so read the [release notes](https://github.com/prismalens/prismalens/releases) before you upgrade.

## What a run looks like

1. An alert arrives from Prometheus Alertmanager, or you write the incident yourself.
2. The agent works in a fresh git worktree of the service’s last commit, never in your checkout.
3. You watch it work on the incident and approve or deny what it asks to do.
4. The report ranks its hypotheses, each with the evidence for and against it, and says what it ruled out and what to check next.

## Where to go next

* [Quick start](/quickstart/): install it and run one investigation, in about ten minutes.
* [Coding agents](/coding-agents/): which agents work, and what they’re allowed to do.
* [Reading the report](/guides/reports/): why hypotheses are ranked, never scored.
* [What PrismaLens reads](/trust/): what the agent can reach, what leaves your machine and where your data lives.

PrismaLens is [Apache-2.0 licensed](https://github.com/prismalens/prismalens/blob/main/LICENSE).