OpenAI Releases Euphony: Visualize AI Agent Workflows and Logs

OpenAI has released Euphony, an open-source, browser-based visualization tool designed to simplify the debugging of complex AI agent workflows. As AI agents increasingly perform multi-step tasks—such as reading files, executing code, and calling APIs—developers have struggled to interpret raw JSON logs that lack the clarity of a traditional stack trace. Euphony addresses this by transforming structured Harmony chat data and Codex session logs into readable, interactive conversation timelines.

Simplifying Complex Data Formats

The tool is specifically engineered to handle the Harmony response format, which is utilized by OpenAI’s gpt-oss model series. Unlike standard chat formats, Harmony supports multi-channel outputs, including reasoning, tool-calling preambles, and role-based instruction hierarchies. Because these conversations often contain deeply nested metadata, token IDs, and rendered display strings, manual inspection is typically difficult. Euphony resolves this by providing a web-based interface that automatically detects and renders these complex structures.
Euphony supports data ingestion via clipboard pasting, local file uploads, or public HTTP(S) URLs, including those from Hugging Face. The tool features an intelligent auto-detection system that identifies whether the input is a list of conversations, a Codex session file, or a custom JSON structure, ensuring that developers can visualize their data without building bespoke log parsers.

Advanced Inspection and Integration

Beyond basic rendering, Euphony offers a robust suite of features for data analysis. Users can leverage JMESPath-based filtering to query large datasets, utilize a focus mode to filter messages by role or content type, and inspect conversation-level metadata through a dedicated panel. The tool also includes a grid view for rapid skimming and an in-browser editor for modifying JSONL content. For content display, it supports markdown rendering, including mathematical formulas and optional HTML.
For development teams looking to integrate these capabilities into their own platforms, Euphony ships as a set of reusable Web Components. These custom elements can be embedded into applications built with React, Svelte, Vue, or plain HTML. Developers can drop a component into their UI and pass data directly through the DOM, with full control over visual styling via CSS custom properties.

Flexible Deployment Options

Euphony is built with a clean architectural split, offering two distinct operating modes. In its frontend-only mode, the application runs entirely within the browser, requiring no server dependency. Alternatively, a backend-assisted mode utilizes a local FastAPI Python server to handle remote loading and backend translation, which is particularly effective for managing large datasets. The project is written primarily in TypeScript and is released under the Apache 2.0 license.