llm-otel-kit

Drop-in OpenTelemetry GenAI observability for any LLM backend — local or cloud.

What it does

llm-otel-kit gives you full OTel GenAI semantic convention coverage for any LLM provider in ~10 lines of code:

• Traces with gen_ai.* span attributes (model, tokens, latency, streaming mode)
• Metrics — 10 instruments: operation duration, token usage, TTFT, TPOT, throughput, error rate, active requests
• Logs exported via OTLP with structured context (model, duration, token counts)
• Dynatrace-ready — correct temporality (DELTA for counters/histograms, CUMULATIVE for UpDownCounters)

Supported Providers

Quick Start

from llm_otel_kit import AppConfig, GenAIMetrics, init_observability, create_provider

config = AppConfig.from_env()
otel = init_observability(config.app_name, config.otlp_endpoint, config.otlp_token)
provider = create_provider(config.provider)
m = GenAIMetrics(otel.meter)

# Use provider.complete() / provider.stream() for instrumented LLM calls

Environment Variables

Install

pip install llm-otel-kit

For Anthropic support:

pip install llm-otel-kit[anthropic]

License

MIT