Docs / Get started
What is Prism
A local tool that answers repository questions no single file can.
Prism reads your repository and answers questions that no single file can answer.
The problem
You join a project, or come back after six months. Your editor shows what any one file says. Nothing tells you which files matter, what a change breaks, which parts belong together as a feature, where tests are thin and churn is high, or what the project is actually built with.
What it does
Prism parses source into a local index and derives views from that one store:
| View | Question |
|---|---|
| Dependency / knowledge / feature graphs | What imports what, what symbols mean, which files implement one capability? |
| Repository DNA | What is this built from, and which domains does it span? |
| Blast radius | If I change this file, what else is affected? |
| Prepare | What should I open for this task, and what already exists? |
| Review | What is Critical, High, Low, or Nit in this change, and what is the Receipt? |
| Lens review | Did Dev, QA, or Architect find a caller left behind, or code no test covers? |
| Findings | What unused or duplicated code is worth fixing? |
| Receipt | Which checks did this change reach, and did they pass? |
Where you use it
| Surface | For |
|---|---|
| MCP server and hooks | Any coding agent: nine tools, a prompt-time pack and a stop-time receipt |
| CLI | Scripts, CI, terminal workflows |
Is it for you?
Probably yes if you work in a codebase too large to hold in your head.
Probably not if the project is small enough to read in an afternoon, or is written in a language Prism does not analyse yet. See known limitations.
Prism writes no code, and its analysis uses no model. Your agent makes the edit, in your chat. Before the edit, Prepare returns the files to open, never the file bodies. When the agent stops, the receipt runs the checks the change reaches and gives a verdict: SHIP, NOT PROVEN or NO-SHIP.
Cost
Disk: an index under ~/.prism/workspaces/ (tens of megabytes on a large
monorepo). Time:
seconds for the first index of a few thousand files, then incremental. Privacy:
nothing leaves your machine. See privacy.