Harvey

Overview

Harvey is a desktop application for qualitative researchers and anyone working with multimedia content. It streamlines the research workflow by integrating a powerful suite of tools into a single, cohesive environment:

• Transcribe audio and video with local AI models.
• Translate transcripts from one language to another.
• Edit and annotate transcripts, documents, PDFs, and images.
• Manage all your project files in one place.

Harvey is named in honor of Sociologist Harvey Sacks (1935-1975), whose foundational work revolutionized social research methods.

Visit the Harvey Website.

Privacy First

Privacy is the core of Harvey's design. All core AI functionalities, like transcription and diarization (speaker identification), run 100% locally on your computer. You download the AI models once and can use them forever offline. Your data never leaves your device.

Developed with AI

Project Harvey has been developed entirely using Vibe Coding to assess the viability of AI-assisted programming in research software development. From its inception, we have developed this application using AI tools such as Gemini, Google AI Studio, and Google Jules. We believe this AI-assisted approach can make development more accessible and efficient, and we encourage contributors to explore and adopt similar methods.

Key Features

• Offline AI Processing:
  • Transcription & Diarization: Use state-of-the-art speech-to-text models (e.g., OpenAI's Whisper) locally to transcribe your audio and video files for complete data privacy.
  • Translation: Translate generated transcripts between multiple languages using local AI models (e.g., NLLB, Helsinki-NLP).
• Interactive Transcript Editor:
  • Rich-text editor linked directly to media playback.
  • Adjust timestamps and manage speaker labels with ease.
• Comprehensive Data Management:
  • Audios and Videos: Import, play, and trim your audio and video files.
  • Documents: Create, edit, and annotate rich-text documents and field notes.
  • PDFs: View and annotate PDF documents with robust text-based tools.
  • Images: Import, view, and annotate images with deep-zoom support.
  • Tables: View and edit CSV and XLSX files with support for nested sub-documents.
• Project & Asset Organization:
  • Manage all your media files, documents, and transcripts across multiple projects.
• User-Friendly & Customizable:
  • Cross-Platform: Works seamlessly on Windows, macOS, and Linux.
  • Configurable: Customize transcription models, application themes, and model download locations.
  • Data Export: Export transcripts to common formats like DOCX and CSV.

Installation Guide

macOS

1. Download the .dmg file from GitHub Releases.
   • For Intel Macs, download the file with x64 in the name.
   • For Apple Silicon (M1, M2, etc.) Macs, download the file with aarch64 in the name.
2. Open the downloaded .dmg file.
3. Drag and drop harvey.app to your /Applications folder.
4. Close the .dmg file window.
5. Open the Terminal application and run the following commands to bypass macOS Gatekeeper:

   cd /Applications
   sudo xattr -dr com.apple.quarantine harvey.app
   

Windows

1. Download the .exe installer file from GitHub Releases.
2. Run the downloaded .exe file.
3. Ensure you have administrative privileges to complete the installation.

Linux

Pre-compiled binaries for Linux are coming soon. In the meantime, you can use our automated bootstrapper to set up your environment and build Harvey from source:

curl -sSL https://raw.githubusercontent.com/Ethnomethodology/harvey/main/scripts/bootstrap.sh | bash

Alternatively, you can clone the repository and run the script manually:

git clone https://github.com/Ethnomethodology/harvey.git
cd harvey && bash scripts/bootstrap.sh

For more detailed manual setup instructions, see our Setup Guide.

Contributing to Harvey

We warmly welcome contributions to Project Harvey from everyone in the open-source community! Whether you're taking your first steps into coding, looking to experiment with AI-assisted development, or a qualitative researcher wanting to shape the tools you use, your input is valuable.

Project Harvey is particularly interested in fostering a collaborative environment that embraces innovative coding paradigms. We encourage contributions from users of "vibe coding" and "agentic coding" platforms (such as those being explored with Google Jules or GitHub Copilot).

Getting Started for Developers & AI Agents

Harvey employs a strict "Visual First" documentation standard.

1. Setup Environment: See Docs/SETUP_GUIDE.md to get your local SvelteKit + Tauri + Rust environment running.
2. Understand the Architecture: Read Docs/PROJECT_DESCRIPTION.md for a high-level overview. For database structure, see Docs/DATABASE_SCHEMA.md.
3. Follow the Rules: Before contributing, you must read CONTRIBUTING.md to understand our documentation templates.
4. For AI Assistants: If you are using an AI agent (like Jules) to modify code, ensure it reads AGENTS.md first to understand its binding constraints regarding documentation updates and styling.

Note: Almost every subdirectory in src/lib/ and src-tauri/src/ contains its own README.md with Mermaid diagrams detailing its specific component architecture and data flow. Always consult these colocated READMEs before modifying code!

Help Center Architecture

The Help Center content is maintained in a Single Source of Truth architecture to ensure consistency between the website and the desktop application.

• Source: All help articles are authored as Markdown files in website/src/content/help.
• Synchronization: A custom script, scripts/sync-help.js, automatically copies these files from the website directory to the desktop app's directory (src/content/help).
• Automation: This synchronization script runs automatically before every dev and build command (via predev and prebuild scripts in package.json), ensuring the desktop app always contains the latest documentation.
• Internal Linking: To ensure compatibility between the website (SvelteKit) and the desktop app (Custom Modal), use relative links for internal navigation (e.g., [Link](slug) or [Link](page-id)). Do not use absolute paths (e.g., /help/slug).

License

This project is licensed under the MIT License.