repo-browser

Local searchable index of all your git repos. Combines FTS5 keyword search with semantic vector search (via Ollama) so you can find repos by name, tag, or plain English description like "something that monitors kubernetes pods".

🌙 Dark mode

☀️ Light mode

---

Contents

• Prerequisites
• Installation and Usage
• CLI
• Configuration
• Files
• Search
• Tag Generation
• Duplicate Detection
• Interactive Duplicate Cleanup
• Deduplication

---

Prerequisites

• Python 3.10+ (stdlib only — no pip installs required)
• Ollama with nomic-embed-text — installed automatically on first run if missing

Platform Support

Platform	Status
Linux	✅ Fully supported
macOS	✅ Fully supported (Homebrew used for Ollama install)
Windows (native)	⚠️ Not supported natively — use WSL (Windows Subsystem for Linux)

On Windows, install WSL and run repo-browser inside it as you would on Linux.

Installation and Usage

Download the latest release tarball from the Releases page and extract it:

cd ~/bin
tar xzf RepoBrowser-1.0.1.tgz
ln -sfn RepoBrowser-1.0.1 RepoBrowser
ln -sf ~/bin/RepoBrowser/repo-browser.py ~/bin/repo-browser.py

The RepoBrowser symlink makes future upgrades a one-liner — just extract the new tarball and update the symlink.

Set up a stable data directory so your database survives upgrades:

mkdir -p ~/.local/share/repo-browser

Create /etc/rb.config:

sudo cp ~/bin/RepoBrowser/rb.config.example /etc/rb.config
sudo vi /etc/rb.config

Set these two values:

gitParent=/path/to/your/git/repos
workDir=/home/youruser/.local/share/repo-browser

Start the server and scan:

repo-browser.py start
repo-browser.py rescan

Open http://localhost:8642 — your repos are now searchable.

Upgrading: extract the new tarball, update the RepoBrowser symlink, and run repo-browser.py rescan. Your database and config are untouched.

CLI

repo-browser.py start     # start server on :8642
repo-browser.py stop      # stop server
repo-browser.py restart   # stop + start
repo-browser.py status    # PID, config source, repo/tag/embed counts
repo-browser.py rescan    # re-scan git dir + re-embed (server stays up)
repo-browser.py duplist   # report duplicate clones to ~/Clone-Duplist.txt
repo-browser.py dupclean  # interactive TUI to delete duplicate clones

Configuration

The config file is read from /etc/rb.config. If not found, the script falls back to rb.config in its own directory. If neither exists, the server starts with defaults and serves the settings UI so you can create one.

# /etc/rb.config
gitParent=/home/user/git
workDir=/home/user/bin/repo-browser

rb.config.example is the template included in the repo. Your local rb.config is gitignored.

Files

File	Purpose
repo-browser.py	Cross-platform launcher — start/stop/restart/status/rescan/duplist/dupclean
scan_repos.py	Walks gitParent, extracts metadata, auto-generates tags, deduplicates by remote URL, populates SQLite
embed_repos.py	Generates semantic embeddings via Ollama nomic-embed-text
repo_search.py	HTTP server on port 8642 — search API + settings API + serves UI
index.html	Single-page frontend with search, tag cloud, settings modal
find-dupe.py	Reports duplicate repo clones across category folders
dupe_clean.py	Curses TUI — interactively select which duplicate clone to delete per repo
rb_config.py	Shared config loader used by all scripts
rb.config.example	Template config (copy to /etc/rb.config)
repos.db	SQLite database (gitignored, regenerated by scan)

Search

Three modes (toggle in the UI header):

• keyword — FTS5 full-text search across repo name, description, README, and tags
• semantic — cosine similarity on nomic-embed-text vectors via Ollama. Finds conceptually related repos even without keyword overlap
• both (default) — weighted blend: 20% keyword + 30% semantic + 20% name match + 30% tag match

Scoring signals:

• Exact name match gets highest priority
• Substring name match (e.g. "MonVisor" finds "MonVisor-Corpus")
• Tag match (searching "rust" prioritises repos tagged rust)
• Semantic-only results are filtered below a 0.65 cosine threshold to reduce noise

Tag Generation

Tags are auto-generated on scan from multiple sources:

• File extensions — .py→python, .go→golang, .rs→rust, .tf→terraform, etc.
• README keywords — matches against ~80 known infra/tool terms (kubernetes, docker, prometheus, ansible, etc.)
• Directory category — parent folder name becomes a tag (e.g. security-automation/)
• Special files — Dockerfile→docker, Jenkinsfile→jenkins, ansible.cfg→ansible

Tags sourced as auto are regenerated on every scan. Manual tags (source manual) are preserved across rescans.

Duplicate Detection

If you organise repos into category folders, the same repo may end up cloned in multiple places. The duplist command finds these and writes a report to ~/Clone-Duplist.txt:

repo-browser.py duplist

Output example:

App Name                     Loc1           Loc2                 Loc3
---------------------------------------------------------------------
bash-textgen                 bash           learning             machine-learning
kubernetes-chatgpt-bot       kubernetes     robusta
WSL-Hello-sudo               WSL            security-automation

Locations are directory names under gitParent. Duplicates are detected by normalised remote URL, with a fallback to name matching for repos without a remote.

Note: the scanner's rescan command also deduplicates automatically — it keeps one copy (the deepest/most specific path) and removes the others from the search index. The duplist report shows you what's on disk. Use dupclean to interactively remove the extras.

Interactive Duplicate Cleanup

dupclean opens a curses TUI that walks you through each duplicated repo one at a time:

repo-browser.py dupclean

For each repo with duplicates you see all clone paths and pick which one to keep. Keys:

Key	Action
↑ / ↓ or k / j	Move cursor
Space	Mark path to keep
Enter / → / n	Confirm selection and proceed to deletion
S	Skip this repo (leave all copies, move on)
Q / Ctrl-C	Quit cleanly

After confirming, you are prompted to confirm each individual deletion before anything is removed. Run repo-browser.py rescan afterwards to update the search index.

Deduplication

Many repos exist in multiple category folders. The scanner deduplicates by:

1. Normalising remote URLs (git@ ↔ https://, trailing .git)
2. Grouping repos with the same URL
3. Keeping the deepest path (most specific category folder)
4. For repos without a remote, deduplicating by name

Stale entries (removed or deduped repos) are cleaned from all tables including FTS5 on every scan.