Stop searching your documents. Start asking them questions.

Arkive is an enterprise-ready RAG (Retrieval-Augmented Generation) knowledge base that lets you upload documents and ask natural language questions — getting accurate, cited answers grounded in your actual files.

🚀 arkive.tianakayemba.dev — no setup required!

Upload your documents and Arkive will:

• Answer questions in natural language — ask anything, get cited answers backed by your documents
• Extract text from PDFs, DOCX, and TXT files — including tables, grade breakdowns, schedules, and structured data
• Cite every answer — each response shows exactly which source and page the information came from
• Filter by document — scope your question to a specific file when you have multiple documents loaded
• Rate confidence — High/Medium/Low confidence badge per answer based on source relevance
• Preview documents — click any file in the library to view its full content with relevant passages highlighted
• Track your session — live stats for documents indexed, chunks stored, queries run, and average relevance
• Copy answers — one-click copy on any AI response for pasting into emails or reports
• Validate uploads — catches password-protected PDFs, empty files, oversized files, and corrupted documents with clear error messages
• Log every query — full observability via Langfuse: latency, token usage, retrieved sources, and answers logged per query

git clone https://github.com/t-skayemba/arkive.git
cd arkive

cd backend
python3.12 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Note: Python 3.12 is required. Python 3.13+ is not yet supported by all dependencies.

cp .env.example .env

Open .env and fill in your keys:

ANTHROPIC_API_KEY=your_anthropic_key_here
LANGFUSE_PUBLIC_KEY=your_langfuse_public_key
LANGFUSE_SECRET_KEY=your_langfuse_secret_key
LANGFUSE_HOST=https://cloud.langfuse.com

Get an Anthropic key at console.anthropic.com Get a free Langfuse account at cloud.langfuse.com

uvicorn main:app --reload --port 8000

cd ../frontend
npm install
npm run dev

Visit http://localhost:5173

1. Upload a document — drag and drop or click the upload zone. Supports PDF, DOCX, TXT up to 20MB
2. Ask a question — type any natural language question and press Enter
3. Filter by document — use the dropdown above the input bar to search within a specific file
4. Review the answer — cited answer with source cards showing which passages were used
5. Preview the document — click any file in the Library to view full text with cited passages highlighted
6. Delete a document — hover over a file in the Library and click the trash icon

All settings are in backend/config.py:

Arkive validates every upload before indexing:

Every query is logged to Langfuse with:

• The question asked and any document filter applied
• Number of chunks retrieved and top relevance score
• The full prompt sent to Claude
• Claude's response
• Input and output token counts
• End-to-end latency in milliseconds

Files and vectors never leave your machine. Query text and relevant document excerpts are sent to Anthropic's API for answer generation. See Anthropic's Privacy Policy for details.

arkive/
├── backend/
│ ├── main.py # FastAPI app entry point
│ ├── config.py # All settings in one place
│ ├── requirements.txt # Python dependencies
│ ├── .env # API keys (not committed)
│ ├── .env.example # Template for required keys
│ ├── routers/
│ │ ├── documents.py # Upload, list, delete, preview endpoints
│ │ └── query.py # Question answering endpoint
│ ├── services/
│ │ ├── document_processor.py # PDF/DOCX/TXT extraction and chunking
│ │ ├── embeddings.py # sentence-transformers embedding service
│ │ └── rag_engine.py # Vector search + Claude + Langfuse logging
│ ├── models/
│ │ └── schemas.py # Pydantic data models
│ └── data/
│ ├── uploads/ # Uploaded document files
│ └── chroma_db/ # ChromaDB vector store
└── frontend/
 └── src/
 ├── App.jsx # Root layout and state
 ├── utils/
 │ └── api.js # Axios API client
 └── components/
 ├── DocumentUpload.jsx # Drag and drop upload with validation
 ├── DocumentLibrary.jsx # File list with delete and preview
 ├── DocumentPreview.jsx # Full text modal with highlights
 ├── QueryInterface.jsx # Chat UI with document filter dropdown
 └── SourceCard.jsx # Citation card component

Interactive docs available at http://localhost:8000/docs when the backend is running.

• Multi-document filtering (search across a selected subset of documents)
• Folder/collection grouping for large document libraries
• Re-index all documents button
• OCR support for scanned PDFs
• Fully local LLM option via Ollama for air-gapped deployments

Built by Tiana Kayemba

MIT