Why RepoBrief?

You need to explain your codebase to an LLM — for code review, debugging, onboarding, or docs. Today, you either:

• Copy-paste files manually — slow, error-prone, easy to forget files or include secrets
• Use a packing tool (Repomix, Gitingest) — great, but they just dump text. No follow-up.
• Use a RAG chatbot (ollama-chat, etc.) — needs ChromaDB, embeddings, GPU, and 16GB+ RAM

RepoBrief does both — packing AND chatting — without the RAM cost:

Installation

pip install repobrief

For cloud LLM support (Claude/GPT):

pip install repobrief[cloud]

That's it. No Docker, no database, no GPU required.

Quick Start

Pack a repo (export mode)

# Pack the current directory into a markdown digest
repobrief pack .

# Save to a file, limit to 50k tokens
repobrief pack ./my-project --max-tokens 50000 -o context.md

# XML format for structured parsing
repobrief pack . --format xml -o context.xml

# Copy directly to clipboard
repobrief pack . --clipboard

# Pack directly from a GitHub URL
repobrief pack https://github.com/pallets/flask

Chat with a repo

# Ask a question using Claude
export ANTHROPIC_API_KEY=sk-ant-...
repobrief chat . --backend cloud -q "Where is user authentication handled?"

# Fully offline using Ollama (no API key needed)
repobrief chat . --backend ollama --model llama3.2:3b -q "What does the payment module do?"

# Interactive mode — ask follow-up questions
repobrief chat . --backend ollama --model llama3.2:3b
> Where is the rate limiter implemented?
> How would I add a new middleware?
> exit

Chat Mode

RepoBrief's chat mode doesn't use RAG, embeddings, or vector databases. Instead, it:

1. Packs your repo into a token-budgeted digest (like export mode)
2. Scores files by relevance to your question (keyword matching, git recency, import centrality)
3. Sends the digest + your question to the chosen LLM backend
4. Streams the answer back to your terminal with real-time markdown rendering

This means it works on any machine that can run Python — even an old laptop with 4GB of free RAM.

Supported Backends

Configuration

CLI Flags

Config File

Create .repobrief.yml in your project root for persistent settings:

backend: ollama
model: llama3.2:3b
max_tokens: 40000
format: markdown
exclude:
  - "*.test.js"
  - "docs/"

Environment Variables

How It Works

repobrief pack/chat <path>
        │
        ▼
  ┌─── SCAN ──────── Walk repo, respect .gitignore, skip binaries
  │
  ├─── SECRET SCAN ── Regex detection of API keys, tokens, passwords
  │
  ├─── SCORE ──────── Rank files by recency, size, centrality, keyword match
  │
  ├─── SELECT ─────── Greedily pick best files within token budget
  │
  ├─── PACK ───────── Build directory tree + concatenated file content
  │
  └─── OUTPUT
        ├── Export: save to file / clipboard / stdout
        └── Chat: send digest + question to Cloud API or local Ollama

Secret Detection

RepoBrief automatically detects and excludes files containing:

• AWS access keys, GitHub PATs, OpenAI/Anthropic API keys
• Private keys (RSA, EC, DSA, OPENSSH)
• .env files with credentials
• Hardcoded passwords and generic secret assignments
• Slack tokens, Stripe keys

Use --redact to include files with secrets replaced by [REDACTED] instead of excluding them entirely.

Development

# Clone and setup
git clone https://github.com/yourusername/repobrief.git
cd repobrief
python -m venv .venv
source .venv/bin/activate  # or .venv\Scripts\activate on Windows
pip install -e ".[dev,cloud]"

# Run tests
pytest

# Lint and format
ruff check src/ tests/
ruff format src/ tests/

# Type check
mypy src/

License

MIT — see LICENSE for details.