zero new SDK

Feed Glimind from OpenTelemetry

If your agents already emit OpenTelemetry GenAI spans (LangChain, Vercel AI, OpenAI Agents, LlamaIndex, or custom), you can get live cross-ecosystem reliability back without adding another SDK. You already produce the data — point it at Glimind.

The one change

export OTEL_EXPORTER_OTLP_TRACES_ENDPOINT=https://glimind.com/v1/traces
export OTEL_EXPORTER_OTLP_PROTOCOL=http/json
# optional — a free key gives higher limits + contribution credit:
export OTEL_EXPORTER_OTLP_HEADERS="authorization=Bearer sk_live_…"

Glimind accepts OTLP/JSON at POST https://glimind.com/v1/traces. We read only the GenAI execute_tool spans and ignore everything else. Prefer a dedicated pipeline? Add Glimind as a second OTLP exporter in your Collector so your primary backend (Datadog, Grafana, Honeycomb…) is untouched.

What we read — and never store

From each execute_tool span we derive, server-side: the tool id, success/failure, latency, value-free input & output shapes, a normalized error class, coarse caller context, and call order within the trace (the workflow graph). The opt-in gen_ai.tool.call.arguments/…result attributes are reduced to value-free shapes on arrival and the raw text is dropped. We never store arguments, results, prompts, or any payload — see privacy.

What you get back

curl https://glimind.com/v1/prepare/github%2Fsearch_repositories?task=search

verdict + recommendation (proceed / caution / avoid)
recipe — a value-free input shape known to work right now
expectedOutput — so you can pre-write parsing
callHints — suggested timeout + retry strategy tuned to this tool
alternatives — live substitutes when it's down

Questions or Collector config help? Docs · free key.