Are OpenTelemetry's GenAI semantic conventions stable enough to build on?

No, not yet. Every gen_ai.* attribute, span, and metric in OpenTelemetry’s GenAI semantic conventions still carries the project’s “development” stability marker, one step short of the “stable” label OpenTelemetry uses once a convention is safe to build production tooling against. The repository has shipped no tagged release, and the spec itself still lists a placeholder where a finished schema URL would go.

That matters most for anyone parsing these fields downstream, not just emitting them. A stable attribute name is a contract: OpenTelemetry won’t restructure it without a deprecation path. A development attribute has no such guarantee, and one already moved, indexed fields like gen_ai.prompt.0.content got replaced by a structured gen_ai.input.messages. A parser written against the old shape stops matching with no error to catch it. Emit the conventions, they’re still the closest thing to a standard, just don’t assume today’s field names survive the next spec revision, and re-check them before every OpenTelemetry upgrade.

sources

keep reading

More on this.

Two ways to run Tessary.

Tessary is an open-source agent reliability platform. Cloud and self-hosted run the same workflow on the OpenTelemetry traces your agent already emits.

Tessary Cloud

We host it for you. Send your first trace with nothing to deploy and no model key.

what's includedper organization
traces
10,000 per calendar month
stored trace data
1 GB
retention
30 days
model credit
$10, one-time, for triage and root-cause analysis
credit card
not required

Self-hosted Tessary

Run the open-source code on your own infrastructure with one command. Add your own model key for triage and root-cause analysis.

Self-host Tessary for me by following https://github.com/tessaryai/tessary/blob/main/setup.md

docker compose -f oci://docker.io/tessaryai/tessary:compose up -d -y