# openai/harmony

Renderer for the harmony response format to be used with gpt-oss

Repository: https://github.com/openai/harmony
Canonical: https://ross.abutalabs.com/products/openai-harmony
Language: Rust
License: Apache-2.0
License Family: permissive
Last push: 2026-04-08T23:52:09+00:00

## Health v2 (maintenance only)
Score: 52/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 76, release rhythm 35, longevity 28
- inputs: {"age_days": 398, "days_push": 147, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 4487, forks 302 (observed 2026-08-28T04:08:51.273422+00:00)

## What it is
OpenAI's Rust library (with first-class Python bindings) for rendering and parsing the harmony response format used by its open-weight gpt-oss models. It handles conversation structure, reasoning channels, and tool-call formatting so custom inference stacks can produce loss-free prompt sequences.

## Use cases
- build a custom inference server for gpt-oss models
- render harmony-format prompts for local LLM inference
- parse model output channels like analysis, commentary, and final
- format tool calls and structured outputs for gpt-oss
- implement gpt-oss support in vLLM or Ollama-like runtimes
- convert chat messages into the harmony token format

## When to choose
- you are building your own inference engine or runtime for gpt-oss
- you need loss-free rendering and parsing of the harmony format
- you want a fast Rust implementation with typed Python bindings

## When to avoid
- you access gpt-oss through an API or provider like HuggingFace, Ollama, or vLLM that already handles formatting
- you work with models that do not use the harmony response format

## Facets
- artifact type: library
- maturity: active
- function: llm-inference, serialization, parser, sdk
- domain: large-language-models, developer-tools
- platform: rust, python, cross-platform
- tags: gpt-oss, prompt-format, tokenizer-format, openai, chat-template, tool-calling, natural-language-processing

## Member repositories
- openai/harmony (main) score 52

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:08:51.273422+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T18:20:30.185803+00:00, confidence not recorded.
  - readme: https://github.com/openai/harmony (fetched 2026-08-28T04:08:51.273422+00:00, sha 31f6d329e841)
- Data as of 2026-08-30T08:39:29.467469+00:00.