← Home

LocalAlibiScan · Open source

Every claim has an alibi.

LocalAlibiScan tells you what every project on your disk is, how it is doing and what changed — and proves each statement with the file and the line it came from. Local, read-only, and it never runs your code.

How it works

Point it at the folder where your projects live and you get one line per project — type, stack, database, external services, last change and alerts:

$ las dashboard ~/code

Then look at one project in depth. Every line is a claim with a status — ✓ confirmed, ≈ inferred, ⚠ contradiction, ? unknown — and below it the evidence: the file, the line and the line itself. Here the README says one thing and the code does another:

$ las scan notes-api --only docs-vs-code --evidence

Documentation vs code
  ⚠ Documentation vs code: Documentation says PostgreSQL, code uses SQLite
      └ README.md:8  - PostgreSQL for persistence
      └ src/db/database.ts:1  import Database from 'better-sqlite3';
      └ package.json:13  "better-sqlite3": "^9.4.0"

The output is in English by default and in Portuguese with las --lang-ui pt — the facts are the same in both.

Why LocalAlibiScan

• Facts with proof. The code extracts the facts — manifests, imports, routes, connection strings. A dependency declared but never used stays “≈ inferred”, not “uses Redis”. When it can't tell, it shows ? instead of guessing.
• The whole disk, not one repository. One dashboard for every project in a folder — including loose script folders without git — and a self-contained HTML report you can open offline or share.
• Documentation that lies. README says PostgreSQL, the code uses SQLite: a ⚠ with both proofs. It never edits anything.
• Changes in architecture language. las refresh says “+ OpenAI service”, “+ New route DELETE /notes/:id”, “⚠ the README doesn't mention OpenAI yet” — not “+42 lines”. It works without git.
• Works without AI. Everything essential is deterministic. A local model (Ollama) is optional: it only writes from the facts found, every sentence must cite them, and a validator removes the ones that don't.
• Terminal or window. Every command works in the terminal; las ui shows the same facts and evidence in a local graphical interface in your browser, listening on this machine only.
• Read-only by construction. It never runs your code, never modifies your files and never goes online. Its only writes go to a .localalibi/ folder that ignores itself in git.

Install

Python 3.11 or later, on macOS, Linux or Windows:

$ pipx install localalibiscan
$ las dashboard ~/code --html

Then las scan for one project, las refresh to see what changed, las ask to ask a question with evidence first. A local model is not required.

Prefer a window? Install the optional graphical interface and open it on the same folder:

$ pipx install 'localalibiscan[ui]'
$ las ui ~/code

The dashboard, every claim with its evidence and the lines around it, what changed, explain and ask — more in the documentation.