Skip to main content
Datadog Agent Observability+32 Positive

GILD Review

GILD's call and the buying facts behind it.

From GILD research

Should we adopt Datadog Agent Observability?

Should a team already running Datadog trace and evaluate its agents in Datadog Agent Observability?

  1. 30-Second TakePick Datadog Agent Observability if your operators work in Datadog and want agent traces under the same trace ID as APM.
  2. Best fitTeams already running Datadog APM that want LLM and agent traces, cost, and evaluations beside service and user data, priced by model call.GILD Review
  3. Consider another workflowIf AI engineers need their own evaluation workspace or a tracer they host, evaluate Arize or Langfuse. If evals gate every release and product teams review them, evaluate Braintrust.GILD Review

Drawn from 63 sources and 10 operator insights

GILD analysis

Datadog Agent Observability Review

Choose Datadog Agent Observability, renamed from LLM Observability in 2026, when your operators work in Datadog and want agent traces under the same trace ID as APM, as Handshake runs it beside Arize AX. It bills only model calls, from US$160 a month for 100,000 LLM spans billed annually, with 40,000 free (Datadog pricing, Oct 28, 2026). Source.ag found duplicate spans overcounting its spend about 2.2 times in June 2026, so check span counts and token totals on your own stack.

  • AI Observability
  • AI Costs

The decision

Best fit
Teams already running Datadog APM that want LLM and agent traces, cost, and evaluations beside service and user data, priced by model call.
Consider another workflow
If AI engineers need their own evaluation workspace or a tracer they host, evaluate Arize or Langfuse. If evals gate every release and product teams review them, evaluate Braintrust.
Adoption condition
Count LLM spans per agent run on a week of real traffic and match token totals against your model bills. Search under both names: Datadog renamed LLM Observability to Agent Observability, and setup still uses llm_observability URLs and DD_LLMOBS settings.

Basis: Datadog pricing and documentation checked Oct 28, 2026, plus 54 public accounts on the Signals page and 11 Operator Insights from named teams, three of them Datadog's own case studies. Most accounts are bug reports filed on Datadog's open-source tracers, which lean toward problems. GILD ran no benchmark of its own. Buying facts rechecked Oct 31, 2026.

Buying & deployment factsWhat a team building on it should verify before production traffic grows.
How you buy it

US$160 a month

Pro billed annually, for the first 100,000 LLM spans a month. Free covers up to 40,000 LLM spans.

Source 1, Datadog, Datadog pricing: Agent Observability: Datadog pricing: Agent Observability

Details

Free covers up to 40,000 LLM spans a month with 15-day trace retention. Pro costs US$160 a month billed annually, US$200 billed month to month, or US$240 billed on demand, for the first 100,000 LLM spans a month.

Cost drivers

US$3.50 per 10,000 spans

LLM spans beyond the included 100,000, billed annually. On demand the rate is US$5.

Source 1, Datadog, Datadog pricing: Agent Observability: Datadog pricing: Agent Observability

Details

The metering unit is the LLM span, a single call to an LLM provider; tool, retrieval, and workflow spans are not billed. LLM spans beyond the included 100,000 cost US$3.50 per 10,000 billed annually, US$4.20 month to month, or US$5 on demand. Trace retention of 30, 60, or 90 days costs US$6.50, US$8, or US$9 per 10,000 LLM spans a month, and Sensitive Data Scanner is included at 1 GB per 10,000 LLM spans.

Where it runs

Datadog SaaS by site

Your organization's Datadog site. Each site is independent, and data is not shared across sites.

Source 2, Datadog, Datadog docs: Getting started with Datadog sites: Datadog docs: Getting started with Datadog sitesSource 3, Datadog, Datadog docs: Agent Observability data governance: Datadog docs: Agent Observability data governance

Details

Datadog runs it as SaaS on the site your organization uses: US1, US3, and US5 in the US, EU1 in Germany, AP1 in Japan, AP2 in Australia, or UK1. Each site is independent, and data is not shared across sites.

What to verify

15-day default retention

Judge evaluations bill tokens to the model account you connect. Longer retention is an add-on.

Source 4, Datadog, Datadog docs: Connect an LLM provider for bring-your-own-key evaluations: Datadog docs: Connect an LLM provider for bring-your-own-key evaluationsSource 3, Datadog, Datadog docs: Agent Observability data governance: Datadog docs: Agent Observability data governanceSource 5, docs.datadoghq.com, Datadog docs: Single Sign On With SAML: Datadog docs: Single Sign On With SAMLSource 6, docs.datadoghq.com, Datadog docs: User Provisioning with SCIM: Datadog docs: User Provisioning with SCIMSource 7, datadoghq.com, Datadog Legal: HIPAA-Eligible Services: Datadog Legal: HIPAA-Eligible ServicesSource 8, docs.datadoghq.com, Datadog docs: Datadog Audit Trail: Datadog docs: Datadog Audit TrailSource 1, Datadog, Datadog pricing: Agent Observability: Datadog pricing: Agent Observability

Details

For judge evaluations you bring your own key, and tokens bill to the account you connect. Traces keep 15 days by default, or 30, 60, or 90 days with a retention add-on. SCIM is available on the Infrastructure Pro, Infrastructure Enterprise, and Startup plans. Datadog lists Agent Observability as a HIPAA Extended Eligible Service under the applicable Business Associate Agreement. Audit Trail is a separate product, from 2% of monthly spend billed annually or 3% on demand.

Sourced from Datadog documentation