🇺🇸 English • 🇨🇳 中文 • 🇹🇼 繁體中文 • 🇯🇵 日本語 • 🇵🇹 Português • 🇧🇷 Português • 🇰🇷 한국어 • 🇪🇸 Español • 🇩🇪 Deutsch • 🇫🇷 Français • 🇮🇱 עברית • 🇸🇦 العربية • 🇷🇺 Русский • 🇵🇱 Polski • 🇨🇿 Čeština • 🇳🇱 Nederlands • 🇹🇷 Türkçe • 🇺🇦 Українська • 🇻🇳 Tiếng Việt • 🇵🇭 Tagalog • 🇮🇩 Indonesia • 🇹🇭 ไทย • 🇮🇳 हिन्दी • 🇧🇩 বাংলা • 🇵🇰 اردو • 🇷🇴 Română • 🇸🇪 Svenska • 🇮🇹 Italiano • 🇬🇷 Ελληνικά…
OpenInference builds on OpenTelemetry to trace AI agents: their LLM calls, tool use,
retrieval and reasoning steps. It is natively supported by Arize Phoenix and
Arize AX, and works with any
OpenTelemetry-compatible backend.
Specification
The OpenInference specification defines how an AI application is recorded as an OpenTelemetry trace: which
span kinds mark LLM calls, agents, tools, retrievers and embeddings, and which attributes hold their inputs, outputs,
token counts and metadata. Every instrumentation follows it, so a trace reads the same whichever framework or language
produced it.
Instrumentation
Instrumentation libraries trace popular model providers and agent frameworks automatically, following the
specification. They are available for Python, JavaScript, Java and Go.
OpenInference Instrumentation for the official OpenAI Go SDK (openai/openai-go).
go get github.com/Arize-ai/openinference/go/openinference-instrumentation-openai-go
Requires Go 1.25+. Pair with arize-otel-go for the one-line OTLP/HTTP setup to Arize AX, or wire up any OTel exporter (e.g. Phoenix at http://localhost:6006/v1/traces).
Supported Destinations
OpenInference traces are standard OpenTelemetry traces, so they can be sent to any backend that accepts OTLP.
Arize Phoenix: open source, with native OpenInference support.
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?v=4)
![dependabot[bot]](https://avatars.githubusercontent.com/in/29110?v=4)
评论