Summary

Euphony tackles the difficulty of inspecting structured chat data from Harmony conversations and Codex sessions. It provides engineers and researchers with portable, customizable, modular Web Components for in-browser visualization, streamlining review across training, evaluation, and agent workflows.

How It Works

Euphony utilizes embeddable Web Components for browser-based rendering of Harmony chat logs and Codex session JSONL files. It supports flexible data loading from clipboard, local files, or public URLs. The core design emphasizes modularity and customization for integration into various web frameworks. It processes data via dedicated viewers for conversations and Codex sessions, featuring Markdown/HTML rendering, metadata inspection, and token visualization.

Quick Start & Requirements

Requires Node.js and pnpm. Build the library with pnpm run build:library. For local development, run the backend (uvicorn fastapi-main:app --app-dir server --host 127.0.0.1 --port 8020 --reload) and frontend (pnpm run dev). The VITE_EUPHONY_FRONTEND_ONLY environment variable controls operational mode. No direct quick-start or demo links are provided in the README.

Highlighted Details

• Dedicated viewers for Harmony conversations and Codex sessions.
• Loads data from clipboard, local files (.json/.jsonl), or HTTP(S) URLs.
• Renders Markdown/HTML content, including formulas.
• Optional translation via user-provided OpenAI API key.
• JMESPath filtering and focus modes for data subsetting.
• Grid and editor modes for dataset skimming and JSONL editing.
• Harmony token rendering displays token IDs, decoded tokens, and display strings.
• Reusable custom elements for integration into React, Svelte, Vue, etc.

Maintenance & Community

The README lacks specific details on maintainers, community channels (Discord/Slack), or project roadmaps.

Licensing & Compatibility

Released under the permissive Apache 2.0 license. A CSS file is MIT licensed (upstream Prism theme). Both licenses permit commercial use and integration into closed-source projects.

Limitations & Caveats

The backend-assisted mode, for local development, poses an SSRF risk if deployed externally due to remote URL fetching. Frontend-only mode requires users to supply their own OpenAI API key for translation.