Why GitIntel

Git already stores everything you need to understand how a codebase evolves. The problem is that the built-in tools answer one question at a time, per invocation, in a format meant for humans reading a single result.

The gap GitIntel fills

Answering "who owns this file and how risky is it?" with plain Git usually looks like this:

git shortlog -sne -- src/gitintel/cli.py     # who touched it
git log --numstat -- src/gitintel/cli.py     # how much churn
git log -1 --format='%an %ad' -- src/...     # who touched it last

Repeat that for every file, aggregate it yourself, and you have the beginnings of an ownership report. GitIntel does that pass once over the whole history and gives you the aggregate in one command:

gitintel ownership . --format json

Design goals

One walk, many reports. The commit walk and per-commit diffs are the expensive part. GitIntel walks history once per run and caches the intermediate results in memory, so summary, contributor, ownership, and hotspot metrics are all derived from the same pass. See Analysis pipeline.

Local and offline by default. Analysis uses libgit2 through pygit2 against the local object database. No GitHub API calls, no rate limits, and no credentials. A GitHub URL is handled by cloning it locally first, then analyzing the clone like any other repository.

Readable in a terminal, parseable in a pipeline. The same analysis renders as a Rich table for humans, JSON for scripts, and Markdown for pull request comments or wiki pages — selected with a single --format flag.

Small surface area. Three commands, four options, no configuration file, no state to manage. The only persistent artifact is the clone cache described in Repository sources and cache.

What GitIntel deliberately is not

• Not a code quality or static analysis tool. It never parses your source code; it only reads commit metadata and diff line counts.
• Not a forge client. It does not read issues, pull requests, reviews, or CI results. Its only remote interaction is a git clone of a public GitHub URL.
• Not a performance dashboard for people. Contributor counts describe the code's history, not individual productivity. See Best practices for how to read the numbers responsibly.
• Not a blame-accurate ownership engine. Ownership is attributed from commit diffs, not from git blame of the current file content. The difference matters and is explained in Metrics and scoring.

Comparison at a glance

Question	Plain Git	GitIntel
Top contributors	git shortlog -sne	gitintel analyze . (with file counts and share)
Churn per file	git log --numstat + manual aggregation	gitintel analyze . (Most Changed Files)
Per-file ownership split	one git log per file	gitintel ownership .
Risk ranking across files	not available	gitintel hotspots .
Machine-readable report	custom formatting flags per command	--format json on every command