What it is
Datadog added LLM observability features to its broader APM platform — capturing traces for OpenAI and Anthropic calls, integrating them with infrastructure metrics, and surfacing them in the same dashboards as the rest of your stack. Pricing is host-based, starting at $15/host/month, with LLM-specific add-ons on top.
Developer experience
If you already have the Datadog agent installed, LLM observability is a few lines of code away. If you don't, you're in for a fully-featured (and fully-priced) APM rollout for the privilege of seeing GPT-4 traces.
Where it shines
- Unified pane of glass. If your CTO insists on one observability vendor, this is the one that already has its hooks in everywhere.
- Alerting. Datadog's alerting and anomaly detection are mature, battle-tested, and miles ahead of what any LLM specialist offers.
- Compliance. Enterprise customers don't have to add a new vendor review.
Where it falls short
- Bolted-on feel. The LLM features clearly post-date the rest of the product. Eval workflows, prompt management, and dataset versioning are either thin or absent.
- Cost. Datadog has never been cheap. Putting LLM traffic through it doesn't make it cheaper.
- Eval depth. You can score traces, but you're a long way from the CI-integrated, dataset-driven workflows that Braintrust or Langfuse offer.
Bottom line
Datadog's LLM observability exists because their customers asked for it, not because anyone thinks it's the best LLM observability product. If you're not already a Datadog shop, don't start now. If you are, it's a reasonable second-best — and it might be the path of least resistance with your security team.