Local App Guidelines

The full script for local version is available on github.com/pharmbio/repuragent. Clone it and run it with Docker. With the Local app, everything will stay on your device, including running processes, input/output files, and conversation history.

2.1 Prerequisites

Requirement

Details

OPENAI_API_KEY

Mandatory. To audit OpenAI models

Docker Desktop

Mandatory. To run Docker Compose. Install Docker Desktop from docker.com.

LangSmith

Optional. To trace the run in detail. More details on smith.langchain.com

2.2 Quick Start

  1. Clone or update the repo

    git clone https://github.com/pharmbio/repuragent.git
cd repuragent

  2. Create .env

    echo "OPENAI_API_KEY=your-openai-api-key-here" > .env

  3. Build and run

    docker-compose up --build

  4. Access the UI
    Visit http://localhost:7860 and follow the Usage Guidelines to start working with the agent.

  5. Stop/restart
    Use docker-compose down to stop. And docker-compose up -d to resume the agent whenever you need.

2.3 Optional settings

LangSmith Tracing

  1. Create a project at smith.langchain.com and generate an API key.

  2. Append to .env:

    echo "LANGCHAIN_TRACING_V2=true"  > .env
echo "LANGCHAIN_ENDPOINT=https://api.smith.langchain.com" > .env
echo "LANGCHAIN_API_KEY=lsm-your-langsmith-key" > .env
echo "LANGCHAIN_PROJECT=your-project-name" > .env

  3. Restart Docker. Every run now emits detailed traces (graph nodes, tool calls, token usage) to the LangSmith dashboard.

2.4 Injecting New SOPs (Local App Only)

The local version lets you inject your own SOP documents into the system.

  1. Drop source files into `data/SOP/ (create SOP directory if needed). The SOP indexer accepts PDF and DOCX; keep filenames descriptive (they appear in citations).

  2. Index or re-index with the provided script:

    cd backend/sop_rag
python sop_indexer.py

    • The script chunks documents, stores summaries in ChromaDB (backend/memory/sop_documents/chromadb/sop_rag), and saves original pages in a docstore folder (backend/memory/sop_documents/docstore).

  3. How it becomes available
    On app startup, SOPRetriever loads the persisted vector store. When an agent calls protocol_search_sop automatically retrieves your newest snippets.

2.5 Local Data Handling

  • Everything is local
    Uploads, results, embeddings, checkpoints, and logs stay under the repo folder you cloned. No data is sent out.

  • No hidden uploads
    The app never transmits your data except to the model endpoints you configure (OpenAI, LangSmith). Review their own privacy terms—the local app relays your prompts/files over the API calls your key authorises.

With the prerequisites in place, you can rely on this guide plus the Usage Guidelines to operate Repuragent locally.