OpenAI Open-Sources Euphony: Advanced Visualization Tool for Harmony and Codex AI Logs

OpenAI Open-Sources Euphony: A Browser-Based Visualization Tool for Harmony Chat Data and Codex Session Logs

OpenAI has officially open-sourced Euphony, a browser-based visualization tool specifically designed to tackle the complexity of debugging multi-step AI agents. The tool transforms raw, nested JSON logs into human-readable conversation timelines, addressing the challenge of reconstructing model reasoning from hundreds of lines of data. This release specifically targets OpenAI’s own Harmony and Codex session formats used in training and evaluation workflows.

Why This Matters

Debugging modern AI agents is fundamentally different from traditional software debugging because there is no single stack trace; instead, developers must parse interleaved token IDs, reasoning outputs, and tool-calling preambles across dozens of steps. Without specialized visualization, engineers are forced to manually inspect raw JSON, which is error-prone and time-consuming during the evaluation of large-scale agentic workflows.

Euphony solves this technical bottleneck by providing a structured UI for multi-channel outputs, allowing teams to see exactly what a model was ‘thinking’ alongside its final response. By supporting role-based instruction hierarchies and named tool namespaces, it provides the necessary granularity to debug complex interactions in models like gpt-oss that typical chat interfaces cannot render.

Key Insights

• Euphony supports the Harmony response format, which allows models like gpt-oss (2026) to generate multi-channel outputs including reasoning and tool calls in a single structure.
• The tool utilizes JMESPath-based filtering, enabling developers to execute complex queries over large JSON datasets to isolate specific message types or roles.
• Built with a TypeScript-heavy codebase (78.7%), the project is released under the Apache 2.0 license, ensuring high compatibility for enterprise integration.
• The application features a dual-operating model: a frontend-only browser mode and a backend-assisted mode powered by a FastAPI Python server for handling remote JSON/JSONL loading.
• Euphony includes specialized detection modes that automatically recognize Codex session files, Harmony lists, or nested conversation fields within arbitrary JSON objects.

Working Examples

Example of embedding the Euphony Web Component into a standard HTML page using a JSON string attribute.

<euphony-conversation
  conversation='{"messages": [{"role": "user", "content": "Hello"}]}'
></euphony-conversation>

Practical Applications

• AI Agent Debugging: Developers can use Focus Mode to filter visible messages by role or content type, allowing them to isolate reasoning failures in multi-step agent logs.
• Dataset Evaluation: Teams can utilize the Grid View to quickly skim through large Harmony-formatted datasets for manual scoring or labeling of model outputs.
• Security Pitfall: OpenAI explicitly warns against exposing the backend-assisted FastAPI server to external networks due to potential Server-Side Request Forgery (SSRF) risks.
• Frontend Integration: Engineers can export Euphony as a reusable custom element to build internal session-replay dashboards within existing React, Svelte, or Vue frameworks.

References:

• https://www.marktechpost.com/2026/04/21/openai-open-sources-euphony-a-browser-based-visualization-tool-for-harmony-chat-data-and-codex-session-logs/