AnswerStack

A project-based RAG application for building knowledge bases from your documents. Upload files, build vector indexes, and ask questions with streamed answers and source citations. Supports speech-to-text and text-to-speech.

Single-user, self-hosted. SQLite for metadata, ChromaDB for vectors, local filesystem for document storage. Works with any OpenAI-compatible LLM provider.

Features

• Create multiple knowledge projects with separate document collections
• Upload PDF, Markdown, and plain text documents
• Automatic chunking with configurable size and overlap
• Vector indexing with incremental rebuild support
• Conversational chat with streamed responses and source citations
• Multi-provider support — configure and switch between LLM providers from the UI
• Configurable retrieval: top-k, similarity threshold, reasoning effort
• FlashRank reranking for improved retrieval quality
• AI-generated suggested prompts based on project content
• Multi-provider text-to-speech: Piper, Kokoro, ElevenLabs
• Speech-to-text via ElevenLabs
• Hardened RAG prompts with injection and hallucination defenses

Prerequisites

• Ubuntu/Debian (or compatible)
• An OpenAI-compatible LLM provider

Quick Start

./scripts/setup.sh       # install dependencies, run migrations, create .env
./scripts/run.sh dev     # backend (port 8000) + vite dev server (port 8080)
./scripts/run.sh prod    # build frontend, serve everything on port 8000

Once running, open the app and configure your LLM provider in Settings.

The setup script installs uv, Python 3.12, and bun if not already present.

Manual Setup

If you prefer to set up each part individually:

# Install all Python dependencies (workspace: backend + core)
uv sync

# Backend
cd backend
uv run alembic upgrade head
uv run uvicorn app.main:app --reload --host 0.0.0.0 --port 8000

# Frontend (in a separate terminal)
cd frontend
bun install
bun run dev

The frontend dev server runs on port 8080 and proxies API requests to the backend on port 8000.

Project Structure

backend/       FastAPI backend, API routes, services, database
core/          answerstack library: parsing, chunking, embedding, retrieval, RAG pipeline
frontend/      React SPA with TypeScript, Vite, shadcn/ui
scripts/       Setup and run scripts

The core/ package is a standalone RAG engine installed as a workspace dependency by the backend. It handles document parsing, text chunking, embedding, vector storage, retrieval with reranking, and the streaming RAG pipeline. All provider communication goes through an OpenAI-compatible abstraction.

Configuration

LLM providers are configured in the Settings page of the UI (provider URL, API key, chat model, embedding model). You can add multiple providers and switch between them.

The .env file handles infrastructure and speech settings:

Variable	Description
DATABASE_URL	SQLite connection string
CHROMADB_PATH	Path for ChromaDB vector storage
DOCUMENT_STORAGE_PATH	Path for uploaded documents
TTS_PROVIDER	Text-to-speech provider: piper, kokoro, or elevenlabs
STT_PROVIDER	Speech-to-text provider: elevenlabs

See .env.example for all options including upload limits and speech provider settings.

Running Tests

# Backend
cd backend
uv run pytest

# Core library
cd core
uv run pytest

# Frontend
cd frontend
bun run test

Tech Stack

Layer	Technology
Backend	FastAPI, SQLAlchemy async, aiosqlite
Database	SQLite with WAL mode
Vector store	ChromaDB
Embeddings	OpenAI-compatible API
Reranking	FlashRank
Frontend	React 18, TypeScript, Vite
UI	shadcn/ui, Radix, Tailwind CSS
State	Zustand, TanStack Query
Package managers	uv, bun

Repo

https://github.com/rwadud/AnswerStack

License

All rights reserved.