repoask 0.2.0

AI-powered repository assistant — ask questions about any codebase

RepoAsk

Ask questions about any codebase using AI. RepoAsk indexes your repository with a RAG pipeline — chunking code by AST symbols, embedding them locally, and answering questions with citations to exact file paths and line numbers.

No server, no database setup. Everything is stored as local files inside your project.

---

Requirements

• Python 3.10+
• An API key for at least one LLM provider (Groq, OpenAI, or Anthropic)

---

Installation

RepoAsk is a CLI tool — install it with pipx so it gets its own isolated environment and is available system-wide.

# Install pipx if you don't have it
brew install pipx
pipx ensurepath

# Install repoask
pipx install repoask

macOS / Homebrew users: Do not use pip install directly — Homebrew manages its Python environment and will block it. pipx is the correct tool for installing Python CLI applications.

Alternative — manual venv (for development or local builds):

python3 -m venv ~/.venvs/repoask
source ~/.venvs/repoask/bin/activate
pip install repoask

---

Quick Start

1. Configure your providers and API keys (once, globally)

repoask init

This walks you through choosing an embedding provider and an LLM provider, then saves your config to ~/.repoask/config.toml.

2. Index a repository

cd /path/to/your/repo
repoask index

RepoAsk scans all Python, JavaScript, and TypeScript files, extracts symbols (functions, classes, interfaces) using tree-sitter, generates embeddings, and stores everything locally under .repoask/.

3. Ask a question

repoask ask "How does authentication work?"

Or start an interactive session:

repoask chat

---

Commands

repoask init

Interactive setup wizard. Sets your embedding provider, LLM provider, models, and API keys. Saves to ~/.repoask/config.toml.

repoask index

Scans and indexes the current repository. Only re-indexes files that have changed since the last run (incremental by default).

repoask index           # incremental (default)
repoask index --full    # force a complete re-index

repoask ask "<question>"

One-shot question. Retrieves relevant code chunks, builds context, and streams an answer.

repoask ask "Where is the database connection initialized?"
repoask ask "What does the UserService class do?"
repoask ask "How are API errors handled?" --top-k 15
repoask ask "Show me all route handlers" --lang typescript

Options:

• --top-k N — number of chunks to retrieve (default: 10)
• --lang LANG — filter results to a specific language (python, javascript, typescript)

repoask chat

Interactive multi-turn session. Maintains conversation context within the session and persists history to disk between sessions.

Commands inside chat:
  /clear    clear the current session context
  /quit     exit

Options:

• --top-k N — number of chunks to retrieve per turn (default: 10)
• --no-history — skip persisting this session to disk

repoask status

Shows index statistics for the current repository.

Repository     /path/to/your/repo
Index path     /path/to/your/repo/.repoask
Embedding      huggingface / sentence-transformers/all-MiniLM-L6-v2
LLM            groq / llama3-8b-8192
Chunks         1,842
Files indexed  74
Last indexed   2026-06-22 14:30:01

repoask config

Displays the current global configuration with API keys masked.

---

Supported Providers

Embedding

Provider	Models	Notes
huggingface	sentence-transformers/all-MiniLM-L6-v2 (default)	Runs locally, no API key needed
openai	text-embedding-3-small, text-embedding-3-large	Requires OpenAI API key

LLM

Provider	Example Models	Notes
groq	llama-3.1-8b-instant, llama-3.3-70b-versatile	Fast, free tier available
openai	gpt-4o-mini, gpt-4o	Requires OpenAI API key
anthropic	claude-haiku-4-5-20251001, claude-sonnet-4-6	Requires Anthropic API key
ollama	llama3.2, mistral, codellama	Fully local, no API key, requires Ollama installed

---

Configuration

Config is stored in TOML format. RepoAsk merges two config files in order:

1. Global ~/.repoask/config.toml — API keys and provider defaults, shared across all repos
2. Per-repo .repoask/config.toml — overrides for a specific project (optional)

Full config reference

[embedding]
provider = "huggingface"
model = "sentence-transformers/all-MiniLM-L6-v2"
api_key = ""

[llm]
provider = "groq"
model = "llama-3.1-8b-instant"
api_key = "gsk_..."
temperature = 0.2
max_tokens = 2048

[indexing]
ignore_patterns = [
    ".git", "node_modules", "__pycache__", "dist", "build",
    ".venv", "venv", "*.lock", "*.min.js", "*.min.css",
    "*.pyc", ".DS_Store", "coverage", ".next", ".nuxt",
]
max_file_size_kb = 500
languages = ["python", "javascript", "typescript"]

[store]
path = ".repoask"

---

Local Index Files

When you run repoask index, a .repoask/ directory is created inside your repository:

.repoask/
├── chroma/        # vector store (ChromaDB, file-based)
├── tracker.db     # file hash registry for incremental indexing (SQLite)
└── history.db     # chat session history (SQLite)

No external database or server is required. Add .repoask/ to your .gitignore to avoid committing the index.

echo ".repoask/" >> .gitignore

---

How It Works

1. Scan — traverses the repository, respecting .gitignore and configured ignore patterns
2. Chunk — parses each file with tree-sitter and extracts symbols (functions, classes, interfaces) as individual chunks instead of fixed token windows
3. Embed — converts each chunk to a vector embedding using your configured provider
4. Store — saves chunks + embeddings + metadata (file path, symbol name, line numbers) to a local ChromaDB collection
5. Retrieve — when you ask a question, it embeds the question and finds the most similar chunks via cosine similarity
6. Build context — enriches the top-K hits with import-referenced definitions and assembles a structured context block
7. Answer — sends the context + question to your LLM with a prompt that requires citations and disallows hallucination

---

Tips

• Run repoask index after pulling new changes — it only re-embeds what changed.
• Use --lang to narrow answers when you know which part of the stack is relevant.
• Increase --top-k for broad architectural questions; lower it for specific function lookups.
• HuggingFace embedding (all-MiniLM-L6-v2) is free and fast enough for most repos. Switch to text-embedding-3-small if you want higher retrieval quality on large codebases.
• Groq is the recommended LLM provider for speed and cost — the free tier is sufficient for most usage.