From e3c17caf422bd3458c96fdee7b8f335a7c563699 Mon Sep 17 00:00:00 2001 From: Greg Svigruha Date: Wed, 2 Sep 2026 14:07:31 -0400 Subject: [PATCH 01/23] Add Agent Observability data retention page Document retention periods for traces, experiment traces, annotated objects, dataset records, prompts, and metrics on a single page. Also correct the annotation queues retention table, which listed annotation labels as retained indefinitely, and fix a broken link to a data retention anchor that did not exist. Co-Authored-By: Claude Opus 5 (1M context) --- hugo/config/_default/menus/main.en.yaml | 9 +- .../en/llm_observability/data_retention.md | 94 +++++++++++++++++++ .../investigate/annotation_queues.md | 8 +- .../llm_observability/investigate/metrics.md | 4 +- 4 files changed, 109 insertions(+), 6 deletions(-) create mode 100644 hugo/content/en/llm_observability/data_retention.md diff --git a/hugo/config/_default/menus/main.en.yaml b/hugo/config/_default/menus/main.en.yaml index 4f7486a03f5..20ba3afa4e5 100644 --- a/hugo/config/_default/menus/main.en.yaml +++ b/hugo/config/_default/menus/main.en.yaml @@ -5784,16 +5784,21 @@ menu: parent: llm_obs identifier: llm_obs_data_security_and_rbac weight: 7 + - name: Data Retention + url: llm_observability/data_retention + parent: llm_obs + identifier: llm_obs_data_retention + weight: 8 - name: Lapdog url: llm_observability/lapdog parent: llm_obs identifier: llm_obs_lapdog - weight: 8 + weight: 9 - name: Guides url: llm_observability/guide/ parent: llm_obs identifier: llm_obs_guide - weight: 9 + weight: 10 - name: GPU Monitoring url: gpu_monitoring/ pre: gpu-monitoring-wui diff --git a/hugo/content/en/llm_observability/data_retention.md b/hugo/content/en/llm_observability/data_retention.md new file mode 100644 index 00000000000..1b257444876 --- /dev/null +++ b/hugo/content/en/llm_observability/data_retention.md @@ -0,0 +1,94 @@ +--- +title: Data Retention +description: Learn how long Agent Observability retains traces, experiments, annotations, datasets, prompts, and metrics. +further_reading: + - link: "/llm_observability/data_privacy_security_and_rbac/" + tag: "Documentation" + text: "Control access to sensitive Agent Observability data" + - link: "/llm_observability/improve/datasets/" + tag: "Documentation" + text: "Work with datasets and dataset versions" + - link: "/llm_observability/investigate/annotation_queues/" + tag: "Documentation" + text: "Review traces with annotation queues" +--- + +Retention periods in Agent Observability depend on the type of data. Traces from your instrumented applications follow the trace retention period in your plan, while experiments, annotations, datasets, and prompts have their own periods. + +## Retention by data type + +| Data | Retention period | +| ------------------------------------------ | ------------------------------------------------------------------- | +| Traces and spans | 15 days by default; 30, 60, or 90 days with a longer-retention plan | +| Experiment traces | 90 days by default; up to 365 days with a longer-retention plan | +| Annotated traces, spans, and sessions | 90 days from the time of annotation, or your trace retention period if that is longer | +| Annotation labels | As long as the object they annotate | +| Dataset records (current version) | 3 years, extended each time the record is used | +| Dataset records (previous versions) | 90 days, extended each time the version is used | +| Prompts in the prompt registry | 3 years, extended each time the prompt is pulled | +| `ml_obs.*` metrics | 15 months | + +## Traces and spans + +Traces and spans from your instrumented applications are retained for **15 days** by default. This applies to everything stored on the span, including per-span operational data such as cost, token counts, latency, and errors, as well as evaluation scores attached to spans. + +Longer retention is available as part of your contract, in tiers of **30, 60, or 90 days**. Because retention affects how much data Datadog stores for you, extending it changes what you are billed for and is arranged with your account team rather than enabled from the Datadog UI. To extend your retention period, contact your Datadog account representative or [Datadog support][1]. + +Your trace retention period also sets the floor for related data. Extending it extends experiment retention and annotation retention along with it, as described in the following sections. + +Retention applies to the raw spans you query in the Traces explorer. Metrics derived from those spans are retained separately, for longer. See [Metrics](#metrics). + +## Experiment traces + +Traces produced by [experiment][3] runs are retained longer than production traces, because experiments are used to compare results across a development cycle. Experiment retention scales with your trace retention period: + +| Trace retention | Experiment trace retention | +| --------------- | -------------------------- | +| 15 days | 90 days | +| 30 days | 180 days | +| 60 days | 270 days | +| 90 days | 365 days | + +Evaluation scores and metrics recorded on experiment rows follow the same period as the experiment traces they are attached to. + +On plans without pre-purchased capacity, experiment traces are retained for the same period as your production traces (15 days) rather than the extended 90 days. + +## Annotated objects + +Annotating an object extends its retention. When you apply an annotation label or note to a trace, span, or session — whether directly or through an [annotation queue][2] — Datadog retains the annotated object for **90 days** from the time of annotation, even if your trace retention period is shorter. Annotating a span retains its whole parent trace, and annotating a trace that belongs to a session retains the whole session. + +If your organization's trace retention period is longer than 90 days, annotated objects are retained for that longer period instead. + +Annotation labels are retained for as long as the object they annotate. After the annotated trace, span, or session expires, its labels are no longer viewable. + +Free-form notes that are not attached to a trace, span, or session are not tied to any object's retention. + +## Dataset records + +[Dataset][4] records are retained for **3 years**. Records in previous versions of a dataset are retained for **90 days**. Both periods are extended each time the record or version is used — for example, when an experiment reads it. After a full retention period with no use, a record becomes eligible for permanent deletion. + +For details on how versions are created, see [Dataset versioning][5]. + +## Prompts + +Prompts in the [prompt registry][9] are retained for **3 years**. This period is extended each time the prompt is pulled by your application, so a prompt in active use stays available. A prompt that is not pulled for 3 years becomes eligible for permanent deletion. + +## Metrics + +The `ml_obs.*` metrics generated from your spans are standard [Datadog metrics][6] and follow [standard Datadog metric retention][7]: 15 months at full granularity. They are retained on this schedule regardless of your trace retention period, so you can build long-term dashboards and monitors on span counts, token usage, cost, latency, and error rates even after the underlying spans expire. + +For the full list of available metrics, see [Agent Observability metrics][8]. + +## Further reading + +{{< partial name="whats-next/whats-next.html" >}} + +[1]: /help/ +[2]: /llm_observability/investigate/annotation_queues/ +[3]: /llm_observability/improve/experiments/ +[4]: /llm_observability/improve/datasets/ +[5]: /llm_observability/improve/datasets/#dataset-versioning +[6]: /metrics/ +[7]: /developers/guide/data-collection-resolution-retention/ +[8]: /llm_observability/investigate/metrics/ +[9]: /llm_observability/configure/prompt_management/ diff --git a/hugo/content/en/llm_observability/investigate/annotation_queues.md b/hugo/content/en/llm_observability/investigate/annotation_queues.md index 80cc2cfce8a..4202052c531 100644 --- a/hugo/content/en/llm_observability/investigate/annotation_queues.md +++ b/hugo/content/en/llm_observability/investigate/annotation_queues.md @@ -325,8 +325,11 @@ You can manage annotation queues programmatically. The following endpoints are a | Data | Retention period | | ----------------- | ----------------------------------------------------| -| Traces in queues | Capped by your organization's trace retention period| -| Annotation labels | Indefinite | +| Traces in queues | Not retained beyond your organization's trace retention period, unless annotated | +| Annotated traces | 90 days from the time of annotation, or your trace retention period if that is longer | +| Annotation labels | As long as the trace they annotate | + +Annotating a trace extends its retention: a trace that would otherwise expire under a shorter trace retention period is retained for 90 days from the time you annotate it. For details, see [Data Retention][16]. ## Example workflows @@ -402,3 +405,4 @@ Build benchmark datasets with human-verified labels for regression testing and c [13]: /api/latest/agent-observability/#get-annotation-queue-label-schema [14]: /api/latest/agent-observability/#update-annotation-queue-label-schema [15]: /account_management/#email-subscriptions +[16]: /llm_observability/data_retention/ diff --git a/hugo/content/en/llm_observability/investigate/metrics.md b/hugo/content/en/llm_observability/investigate/metrics.md index e223e31bb7b..891b639ef0d 100644 --- a/hugo/content/en/llm_observability/investigate/metrics.md +++ b/hugo/content/en/llm_observability/investigate/metrics.md @@ -25,8 +25,8 @@ The ml_obs.* entries on this page are Datadog M

They are distinct from two other things in Agent Observability: From d69e761d186cf8d62e098db57a0bc1b18d6b7c1d Mon Sep 17 00:00:00 2001 From: Greg Svigruha Date: Wed, 2 Sep 2026 14:30:55 -0400 Subject: [PATCH 02/23] Correct dataset record retention on data retention page The 3-year period for records in the current version runs from creation and is not extended by reads. Only the 90-day period for previous versions resets when a version is used. Co-Authored-By: Claude Opus 5 (1M context) --- hugo/content/en/llm_observability/data_retention.md | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) diff --git a/hugo/content/en/llm_observability/data_retention.md b/hugo/content/en/llm_observability/data_retention.md index 1b257444876..78fc96d634c 100644 --- a/hugo/content/en/llm_observability/data_retention.md +++ b/hugo/content/en/llm_observability/data_retention.md @@ -23,7 +23,7 @@ Retention periods in Agent Observability depend on the type of data. Traces from | Experiment traces | 90 days by default; up to 365 days with a longer-retention plan | | Annotated traces, spans, and sessions | 90 days from the time of annotation, or your trace retention period if that is longer | | Annotation labels | As long as the object they annotate | -| Dataset records (current version) | 3 years, extended each time the record is used | +| Dataset records (current version) | 3 years from creation | | Dataset records (previous versions) | 90 days, extended each time the version is used | | Prompts in the prompt registry | 3 years, extended each time the prompt is pulled | | `ml_obs.*` metrics | 15 months | @@ -65,7 +65,9 @@ Free-form notes that are not attached to a trace, span, or session are not tied ## Dataset records -[Dataset][4] records are retained for **3 years**. Records in previous versions of a dataset are retained for **90 days**. Both periods are extended each time the record or version is used — for example, when an experiment reads it. After a full retention period with no use, a record becomes eligible for permanent deletion. +Records in the current version of a [dataset][4] are retained for **3 years** from the time they are created. + +Records in previous versions of a dataset are retained for **90 days**. This period is extended each time a previous version is used — for example, when an experiment reads that version. After 90 consecutive days without use, a previous version becomes eligible for permanent deletion. For details on how versions are created, see [Dataset versioning][5]. From d9d5464fec0de7601ff4e625301a4460c69b2284 Mon Sep 17 00:00:00 2001 From: Greg Svigruha Date: Wed, 2 Sep 2026 14:33:35 -0400 Subject: [PATCH 03/23] Correct experiment trace retention on data retention page Experiment retention pairs with the trace retention tier only on plans with pre-purchased capacity. Without it, experiment traces are retained for 15 days rather than 90. Note that custom contracts may not follow the tier pairings. Co-Authored-By: Claude Opus 5 (1M context) --- hugo/content/en/llm_observability/data_retention.md | 10 ++++++---- 1 file changed, 6 insertions(+), 4 deletions(-) diff --git a/hugo/content/en/llm_observability/data_retention.md b/hugo/content/en/llm_observability/data_retention.md index 78fc96d634c..970efd50b44 100644 --- a/hugo/content/en/llm_observability/data_retention.md +++ b/hugo/content/en/llm_observability/data_retention.md @@ -20,7 +20,7 @@ Retention periods in Agent Observability depend on the type of data. Traces from | Data | Retention period | | ------------------------------------------ | ------------------------------------------------------------------- | | Traces and spans | 15 days by default; 30, 60, or 90 days with a longer-retention plan | -| Experiment traces | 90 days by default; up to 365 days with a longer-retention plan | +| Experiment traces | 15 days without pre-purchased capacity; 90 to 365 days with it | | Annotated traces, spans, and sessions | 90 days from the time of annotation, or your trace retention period if that is longer | | Annotation labels | As long as the object they annotate | | Dataset records (current version) | 3 years from creation | @@ -34,13 +34,13 @@ Traces and spans from your instrumented applications are retained for **15 days* Longer retention is available as part of your contract, in tiers of **30, 60, or 90 days**. Because retention affects how much data Datadog stores for you, extending it changes what you are billed for and is arranged with your account team rather than enabled from the Datadog UI. To extend your retention period, contact your Datadog account representative or [Datadog support][1]. -Your trace retention period also sets the floor for related data. Extending it extends experiment retention and annotation retention along with it, as described in the following sections. +Your trace retention period also affects related data. On standard plans, a longer trace retention tier comes with a longer experiment retention period, and trace retention sets the floor for how long annotated objects are kept. See the following sections. Retention applies to the raw spans you query in the Traces explorer. Metrics derived from those spans are retained separately, for longer. See [Metrics](#metrics). ## Experiment traces -Traces produced by [experiment][3] runs are retained longer than production traces, because experiments are used to compare results across a development cycle. Experiment retention scales with your trace retention period: +Traces produced by [experiment][3] runs are retained longer than production traces on plans with pre-purchased capacity, because experiments are used to compare results across a development cycle. On these plans, each trace retention tier comes with a longer experiment retention period: | Trace retention | Experiment trace retention | | --------------- | -------------------------- | @@ -49,9 +49,11 @@ Traces produced by [experiment][3] runs are retained longer than production trac | 60 days | 270 days | | 90 days | 365 days | +Without pre-purchased capacity — including organizations that have no Agent Observability entitlement — experiment traces are retained for the same **15 days** as production traces. + Evaluation scores and metrics recorded on experiment rows follow the same period as the experiment traces they are attached to. -On plans without pre-purchased capacity, experiment traces are retained for the same period as your production traces (15 days) rather than the extended 90 days. +If your organization has a custom contract, your retention periods may not follow the pairings in this table. Contact your Datadog account representative to confirm your periods. ## Annotated objects From 88657a73a41ad472edb652391cea1aaa36ca8e47 Mon Sep 17 00:00:00 2001 From: Greg Svigruha Date: Wed, 2 Sep 2026 14:38:04 -0400 Subject: [PATCH 04/23] Add experiment definition retention, soften annotation label claim Experiment definitions and aggregate results carry a flat 90-day period that does not scale with the trace retention tier, unlike experiment traces. State annotation label availability in terms of the annotated object rather than asserting label deletion. Co-Authored-By: Claude Opus 5 (1M context) --- hugo/content/en/llm_observability/data_retention.md | 12 +++++++----- 1 file changed, 7 insertions(+), 5 deletions(-) diff --git a/hugo/content/en/llm_observability/data_retention.md b/hugo/content/en/llm_observability/data_retention.md index 970efd50b44..d8f2e1de8dc 100644 --- a/hugo/content/en/llm_observability/data_retention.md +++ b/hugo/content/en/llm_observability/data_retention.md @@ -21,8 +21,8 @@ Retention periods in Agent Observability depend on the type of data. Traces from | ------------------------------------------ | ------------------------------------------------------------------- | | Traces and spans | 15 days by default; 30, 60, or 90 days with a longer-retention plan | | Experiment traces | 15 days without pre-purchased capacity; 90 to 365 days with it | +| Experiment definitions and aggregate results | 90 days from creation, on all plans | | Annotated traces, spans, and sessions | 90 days from the time of annotation, or your trace retention period if that is longer | -| Annotation labels | As long as the object they annotate | | Dataset records (current version) | 3 years from creation | | Dataset records (previous versions) | 90 days, extended each time the version is used | | Prompts in the prompt registry | 3 years, extended each time the prompt is pulled | @@ -51,19 +51,21 @@ Traces produced by [experiment][3] runs are retained longer than production trac Without pre-purchased capacity — including organizations that have no Agent Observability entitlement — experiment traces are retained for the same **15 days** as production traces. -Evaluation scores and metrics recorded on experiment rows follow the same period as the experiment traces they are attached to. - If your organization has a custom contract, your retention periods may not follow the pairings in this table. Contact your Datadog account representative to confirm your periods. +### Experiment definitions + +The experiment itself — its name, configuration, and aggregate results — is retained for **90 days** from the time it is created. Unlike experiment traces, this period is the same on every plan and does not extend with a longer trace retention tier. Export any experiment results you need to keep beyond 90 days. + ## Annotated objects Annotating an object extends its retention. When you apply an annotation label or note to a trace, span, or session — whether directly or through an [annotation queue][2] — Datadog retains the annotated object for **90 days** from the time of annotation, even if your trace retention period is shorter. Annotating a span retains its whole parent trace, and annotating a trace that belongs to a session retains the whole session. If your organization's trace retention period is longer than 90 days, annotated objects are retained for that longer period instead. -Annotation labels are retained for as long as the object they annotate. After the annotated trace, span, or session expires, its labels are no longer viewable. +Annotation labels are viewable for as long as the object they annotate is retained. -Free-form notes that are not attached to a trace, span, or session are not tied to any object's retention. +Free-form notes are not attached to a trace, span, or session, so no object's retention applies to them. ## Dataset records From da3f3080c500be78b28311e8e00a91456847e013 Mon Sep 17 00:00:00 2001 From: Greg Svigruha Date: Wed, 2 Sep 2026 16:11:03 -0400 Subject: [PATCH 05/23] Add Agent Observability billing page, scope retention page to retention Add a billing page covering how Agent Observability is metered, what counts as an LLM span, usage estimation from provider request counts, and the Sensitive Data Scanner allotment. Rates link to the pricing page rather than being restated. Scope the data retention page to retention periods, moving the commercial detail to the billing page, and correct it against the current plan structure: retention is keyed on plan rather than on the trace retention period, and retention changes apply retroactively to unexpired spans. Co-Authored-By: Claude Opus 5 (1M context) --- hugo/config/_default/menus/main.en.yaml | 5 + .../en/account_management/billing/_index.md | 1 + .../billing/llm_observability.md | 99 +++++++++++++++++++ .../en/llm_observability/data_retention.md | 69 ++++++------- 4 files changed, 141 insertions(+), 33 deletions(-) create mode 100644 hugo/content/en/account_management/billing/llm_observability.md diff --git a/hugo/config/_default/menus/main.en.yaml b/hugo/config/_default/menus/main.en.yaml index 20ba3afa4e5..2bbf0493fc1 100644 --- a/hugo/config/_default/menus/main.en.yaml +++ b/hugo/config/_default/menus/main.en.yaml @@ -10720,6 +10720,11 @@ menu: url: account_management/billing/ai_credit_limits/ parent: account_management_billing weight: 1522 + - name: Agent Observability + identifier: account_management_billing_llm_observability + url: account_management/billing/llm_observability/ + parent: account_management_billing + weight: 1523 - name: Multi-org Accounts identifier: account_management_multi_org_accounts url: account_management/multi_organization/ diff --git a/hugo/content/en/account_management/billing/_index.md b/hugo/content/en/account_management/billing/_index.md index bcfb8eed104..fe90063b1ff 100644 --- a/hugo/content/en/account_management/billing/_index.md +++ b/hugo/content/en/account_management/billing/_index.md @@ -108,6 +108,7 @@ If you pay by check, ACH, or wire, invoices are emailed to the billing email add {{< nextlink href="account_management/billing/serverless/" >}}Serverless{{< /nextlink >}} {{< nextlink href="account_management/billing/rum/" >}}Real User Monitoring{{< /nextlink >}} {{< nextlink href="account_management/billing/ci_visibility/" >}}CI Visibility{{< /nextlink >}} + {{< nextlink href="account_management/billing/llm_observability/" >}}Agent Observability{{< /nextlink >}} {{< nextlink href="account_management/billing/aws/" >}}AWS integration{{< /nextlink >}} {{< nextlink href="account_management/billing/azure/" >}}Azure integration{{< /nextlink >}} {{< nextlink href="account_management/billing/alibaba/" >}}Alibaba integration{{< /nextlink >}} diff --git a/hugo/content/en/account_management/billing/llm_observability.md b/hugo/content/en/account_management/billing/llm_observability.md new file mode 100644 index 00000000000..46cb17b2329 --- /dev/null +++ b/hugo/content/en/account_management/billing/llm_observability.md @@ -0,0 +1,99 @@ +--- +title: Agent Observability Billing +algolia: + tags: ['agent observability billing', 'llm observability billing', 'agent observability pricing', 'llm observability pricing', 'llm spans'] +further_reading: +- link: "/llm_observability/" + tag: "Documentation" + text: "Learn about Agent Observability" +- link: "/llm_observability/data_retention/" + tag: "Documentation" + text: "Learn how long Agent Observability retains your data" +--- + +## Overview + +This page contains common questions and answers about billing topics for [Agent Observability][1]. + +## How is Agent Observability billed? + +Agent Observability is metered and billed on the number of LLM spans ingested. For current rates, see the [Datadog pricing page][2]. + +## What counts as an LLM span? + +An LLM span represents a single request to an LLM provider, such as OpenAI or Anthropic. + +One agent workflow can make several LLM requests, and each one produces its own LLM span. A trace containing seven LLM requests therefore contains seven billable LLM spans. Datadog bills only on LLM spans, not on the surrounding workflow, tool, and retrieval spans in the same trace. + +## Are experiments billed separately? + +No. [Experiments][3] are part of Agent Observability and are billed the same way as production monitoring, on the number of LLM spans ingested. + +## Are indexed spans billed separately? + +No. Indexed LLM spans are included in the base price. + +## Is there a free allotment of LLM spans? + +Yes. Each organization receives an allotment of LLM spans per month at no cost. For the current allotment, see the [Datadog pricing page][2]. + +## How can I estimate my LLM span usage? + +Each request your application makes to an LLM provider corresponds to one LLM span, so provider-side request counts are a good starting point. Most providers report request counts in their own console, and Datadog integrations expose the same counts as metrics: + +| Provider | Where to find request counts | +| ------------------------------- | --------------------------------------------------------------------------------------------- | +| OpenAI | Total requests in the OpenAI admin UI, or sum the `*.num_model_requests` metrics | +| AWS Bedrock | Invocations by model in CloudWatch, or the `aws.bedrock.invocations` metric | +| Google Enterprise Agent Platform | Usage metrics in the platform console, or the `gcp.aiplatform.online_prediction_requests_per_base_model` metric | +| Vercel | Requests by provider and model in the Vercel dashboard, or the `vercel.requests` metric | +| Anthropic | Request totals are not reported; use your own application metrics as a proxy | + +Keep three things in mind when using provider request counts: + +- They are closer to an upper bound than an exact figure, because not every request is instrumented as an LLM span. +- If you [sample][4] your traces, your billable spans are a fraction of your total requests. +- Counting inbound requests to your own application undercounts if each one triggers several LLM calls. An agent that makes five LLM calls per user query produces five times the spans. + +If your application uses a supported provider, the Agent Observability SDK captures these requests automatically. See the [Python SDK][5] and [Node.js SDK][6] documentation for supported providers. + +## How can I see my billable usage? + +In Datadog, go to [{{< ui >}}Plan & Usage{{< /ui >}}][7] and open the {{< ui >}}Agent Observability{{< /ui >}} tab. + +## Does extending my data retention period affect billing? + +Yes. Longer retention is a paid add-on, because it increases how much data Datadog stores for you. Add-ons extend span retention to 30, 60, or 90 days, along with a correspondingly longer period for experiment traces. For the periods each add-on provides, see [Data Retention][8]. To add one, contact your Datadog account representative or [Datadog support][9]. + +## Is Agent Observability a standalone product? + +Yes. Agent Observability does not require you to purchase any other Datadog product. + +## Is Sensitive Data Scanner included? + +Agent Observability uses [Sensitive Data Scanner][10] Library Rules to identify and redact sensitive information in your LLM application traffic, including personal information, financial data, and health records. You do not purchase Sensitive Data Scanner separately to use this capability. + +Your Sensitive Data Scanner allotment scales with your LLM span usage: every 10,000 LLM spans includes 1 GB of Sensitive Data Scanner usage. + +| LLM spans | Included Sensitive Data Scanner usage | +| --------- | ------------------------------------- | +| 40,000 | 4 GB | +| 100,000 | 10 GB | +| 300,000 | 30 GB | + +This allotment applies to every plan. Usage beyond your allotment is billed as Sensitive Data Scanner usage. + +## Further reading + +{{< partial name="whats-next/whats-next.html" >}} + +[1]: /llm_observability/ +[2]: https://www.datadoghq.com/pricing/ +[3]: /llm_observability/improve/experiments/ +[4]: /llm_observability/instrument/sdk/ +[5]: /llm_observability/instrument/agentic/python +[6]: /llm_observability/instrument/agentic/nodejs +[7]: https://app.datadoghq.com/billing/usage +[8]: /llm_observability/data_retention/ +[9]: /help/ +[10]: /security/sensitive_data_scanner/ diff --git a/hugo/content/en/llm_observability/data_retention.md b/hugo/content/en/llm_observability/data_retention.md index d8f2e1de8dc..51f4a40dd39 100644 --- a/hugo/content/en/llm_observability/data_retention.md +++ b/hugo/content/en/llm_observability/data_retention.md @@ -13,55 +13,59 @@ further_reading: text: "Review traces with annotation queues" --- -Retention periods in Agent Observability depend on the type of data. Traces from your instrumented applications follow the trace retention period in your plan, while experiments, annotations, datasets, and prompts have their own periods. +Retention periods in Agent Observability depend on the type of data and on your plan. Traces from your instrumented applications follow the span retention period in your plan, while experiment definitions, datasets, and prompts have their own periods. ## Retention by data type -| Data | Retention period | -| ------------------------------------------ | ------------------------------------------------------------------- | -| Traces and spans | 15 days by default; 30, 60, or 90 days with a longer-retention plan | -| Experiment traces | 15 days without pre-purchased capacity; 90 to 365 days with it | -| Experiment definitions and aggregate results | 90 days from creation, on all plans | -| Annotated traces, spans, and sessions | 90 days from the time of annotation, or your trace retention period if that is longer | -| Dataset records (current version) | 3 years from creation | -| Dataset records (previous versions) | 90 days, extended each time the version is used | -| Prompts in the prompt registry | 3 years, extended each time the prompt is pulled | -| `ml_obs.*` metrics | 15 months | +| Data | Retention period | +| -------------------------------------------- | ----------------------------------------------------------------------------------------- | +| Traces and spans | 15 days; 30, 60, or 90 days with a retention add-on | +| Experiment traces | 15 days on on-demand plans; 90 days on committed plans; 6, 9, or 12 months with a retention add-on | +| Experiment definitions and aggregate results | 90 days from creation | +| Annotated traces, spans, and sessions | 90 days from the time of annotation, or your span retention period if that is longer | +| Dataset records | 3 years, regardless of your span retention period | +| Prompts in the prompt registry | 3 years, extended each time the prompt is pulled | +| `ml_obs.*` metrics | 15 months | ## Traces and spans -Traces and spans from your instrumented applications are retained for **15 days** by default. This applies to everything stored on the span, including per-span operational data such as cost, token counts, latency, and errors, as well as evaluation scores attached to spans. +Traces and spans from your instrumented applications are retained for **15 days** on all plans by default. This applies to everything stored on the span, including per-span operational data such as cost, token counts, latency, and errors, as well as evaluation scores attached to spans. -Longer retention is available as part of your contract, in tiers of **30, 60, or 90 days**. Because retention affects how much data Datadog stores for you, extending it changes what you are billed for and is arranged with your account team rather than enabled from the Datadog UI. To extend your retention period, contact your Datadog account representative or [Datadog support][1]. - -Your trace retention period also affects related data. On standard plans, a longer trace retention tier comes with a longer experiment retention period, and trace retention sets the floor for how long annotated objects are kept. See the following sections. +A retention add-on extends this to **30, 60, or 90 days**. To request a longer retention period, contact your Datadog account representative or [Datadog support][1]. For how retention affects billing, see [Agent Observability Billing][10]. Retention applies to the raw spans you query in the Traces explorer. Metrics derived from those spans are retained separately, for longer. See [Metrics](#metrics). ## Experiment traces -Traces produced by [experiment][3] runs are retained longer than production traces on plans with pre-purchased capacity, because experiments are used to compare results across a development cycle. On these plans, each trace retention tier comes with a longer experiment retention period: - -| Trace retention | Experiment trace retention | -| --------------- | -------------------------- | -| 15 days | 90 days | -| 30 days | 180 days | -| 60 days | 270 days | -| 90 days | 365 days | +Traces produced by [experiment][3] runs are retained longer than production traces on committed plans. -Without pre-purchased capacity — including organizations that have no Agent Observability entitlement — experiment traces are retained for the same **15 days** as production traces. +| Plan | Experiment trace retention | +| ----------------------------------- | -------------------------- | +| On-demand | 15 days | +| Committed (monthly or annual) | 90 days | +| 30-day retention add-on | 6 months | +| 60-day retention add-on | 9 months | +| 90-day retention add-on | 12 months | -If your organization has a custom contract, your retention periods may not follow the pairings in this table. Contact your Datadog account representative to confirm your periods. +If your organization has a custom contract, your retention periods may not match this table. Contact your Datadog account representative to confirm your periods. ### Experiment definitions -The experiment itself — its name, configuration, and aggregate results — is retained for **90 days** from the time it is created. Unlike experiment traces, this period is the same on every plan and does not extend with a longer trace retention tier. Export any experiment results you need to keep beyond 90 days. +The experiment itself — its name, configuration, and aggregate results — is retained for **90 days** from the time it is created. Unlike experiment traces, this period is the same on every plan and does not extend with a retention add-on. Export any experiment results you need to keep beyond 90 days. + +## Changing your retention period + +When you add or extend a retention add-on, the longer period applies **retroactively to every span that has not already expired**. Spans that expired under your previous period are not recoverable. + +For example, if you are on the default 15-day retention and add a 60-day add-on today, the spans from the last 15 days pick up the 60-day period, but anything older is already gone. + +When you move to a shorter retention period, spans older than the new period are no longer available. ## Annotated objects -Annotating an object extends its retention. When you apply an annotation label or note to a trace, span, or session — whether directly or through an [annotation queue][2] — Datadog retains the annotated object for **90 days** from the time of annotation, even if your trace retention period is shorter. Annotating a span retains its whole parent trace, and annotating a trace that belongs to a session retains the whole session. +Annotating an object extends its retention. When you apply an annotation label or note to a trace, span, or session — whether directly or through an [annotation queue][2] — Datadog retains the annotated object for **90 days** from the time of annotation, even if your span retention period is shorter. Annotating a span retains its whole parent trace, and annotating a trace that belongs to a session retains the whole session. -If your organization's trace retention period is longer than 90 days, annotated objects are retained for that longer period instead. +If your organization's span retention period is longer than 90 days, annotated objects are retained for that longer period instead. Annotation labels are viewable for as long as the object they annotate is retained. @@ -69,11 +73,9 @@ Free-form notes are not attached to a trace, span, or session, so no object's re ## Dataset records -Records in the current version of a [dataset][4] are retained for **3 years** from the time they are created. - -Records in previous versions of a dataset are retained for **90 days**. This period is extended each time a previous version is used — for example, when an experiment reads that version. After 90 consecutive days without use, a previous version becomes eligible for permanent deletion. +Records in a [dataset][4] are retained for **3 years**, regardless of your span retention period. -For details on how versions are created, see [Dataset versioning][5]. +Records in previous versions of a dataset are retained for **90 days**. This period is extended each time a previous version is used — for example, when an experiment reads that version. After 90 consecutive days without use, a previous version becomes eligible for permanent deletion. For details on how versions are created, see [Dataset versioning][5]. ## Prompts @@ -81,7 +83,7 @@ Prompts in the [prompt registry][9] are retained for **3 years**. This period is ## Metrics -The `ml_obs.*` metrics generated from your spans are standard [Datadog metrics][6] and follow [standard Datadog metric retention][7]: 15 months at full granularity. They are retained on this schedule regardless of your trace retention period, so you can build long-term dashboards and monitors on span counts, token usage, cost, latency, and error rates even after the underlying spans expire. +The `ml_obs.*` metrics generated from your spans are standard [Datadog metrics][6] and follow [standard Datadog metric retention][7]: 15 months at full granularity. They are retained on this schedule regardless of your span retention period, so you can build long-term dashboards and monitors on span counts, token usage, cost, latency, and error rates even after the underlying spans expire. For the full list of available metrics, see [Agent Observability metrics][8]. @@ -98,3 +100,4 @@ For the full list of available metrics, see [Agent Observability metrics][8]. [7]: /developers/guide/data-collection-resolution-retention/ [8]: /llm_observability/investigate/metrics/ [9]: /llm_observability/configure/prompt_management/ +[10]: /account_management/billing/llm_observability/ From 3f046ebe419a0caaa0e854d37e2566af322d0f5b Mon Sep 17 00:00:00 2001 From: Greg Svigruha Date: Wed, 2 Sep 2026 16:13:37 -0400 Subject: [PATCH 06/23] Move Agent Observability billing entry into the product group Place it after Incident Response at the end of the per-product billing pages, before the cloud integrations, matching its position on the billing index. Renumber the following entries to keep weights contiguous. Co-Authored-By: Claude Opus 5 (1M context) --- hugo/config/_default/menus/main.en.yaml | 26 ++++++++++++------------- 1 file changed, 13 insertions(+), 13 deletions(-) diff --git a/hugo/config/_default/menus/main.en.yaml b/hugo/config/_default/menus/main.en.yaml index 2bbf0493fc1..1592c69a142 100644 --- a/hugo/config/_default/menus/main.en.yaml +++ b/hugo/config/_default/menus/main.en.yaml @@ -10675,55 +10675,55 @@ menu: url: account_management/billing/incident_response/ parent: account_management_billing weight: 1513 + - name: Agent Observability + identifier: account_management_billing_llm_observability + url: account_management/billing/llm_observability/ + parent: account_management_billing + weight: 1514 - name: AWS Integration identifier: account_management_billing_aws url: account_management/billing/aws/ parent: account_management_billing - weight: 1514 + weight: 1515 - name: Azure Integration identifier: account_management_billing_azure url: account_management/billing/azure/ parent: account_management_billing - weight: 1515 + weight: 1516 - name: Google Cloud Integration identifier: account_management_billing_google_cloud url: account_management/billing/google_cloud/ parent: account_management_billing - weight: 1516 + weight: 1517 - name: OCI Integration identifier: account_management_billing_oci url: account_management/billing/oci/ parent: account_management_billing - weight: 1517 + weight: 1518 - name: Alibaba Integration identifier: account_management_billing_alibaba url: account_management/billing/alibaba/ parent: account_management_billing - weight: 1518 + weight: 1519 - name: vSphere Integration identifier: account_management_billing_vsphere url: account_management/billing/vsphere/ parent: account_management_billing - weight: 1519 + weight: 1520 - name: Workflow Automation identifier: account_management_billing_workflow_automation url: account_management/billing/workflow_automation/ parent: account_management_billing - weight: 1520 + weight: 1521 - name: AI Credits identifier: account_management_billing_ai_credits url: account_management/billing/ai_credits/ parent: account_management_billing - weight: 1521 + weight: 1522 - name: AI Credit Limits identifier: account_management_billing_ai_credit_limits url: account_management/billing/ai_credit_limits/ parent: account_management_billing - weight: 1522 - - name: Agent Observability - identifier: account_management_billing_llm_observability - url: account_management/billing/llm_observability/ - parent: account_management_billing weight: 1523 - name: Multi-org Accounts identifier: account_management_multi_org_accounts From 7151ee3d652db9ffd1ff4c5b9540eaec6f0b1fcb Mon Sep 17 00:00:00 2001 From: Greg Svigruha Date: Wed, 2 Sep 2026 16:19:49 -0400 Subject: [PATCH 07/23] Use Google Vertex AI in the usage estimation table Match the name used throughout the docs and parallel the other provider rows, instead of introducing a name used nowhere else. Co-Authored-By: Claude Opus 5 (1M context) --- hugo/content/en/account_management/billing/llm_observability.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/hugo/content/en/account_management/billing/llm_observability.md b/hugo/content/en/account_management/billing/llm_observability.md index 46cb17b2329..49258a321d0 100644 --- a/hugo/content/en/account_management/billing/llm_observability.md +++ b/hugo/content/en/account_management/billing/llm_observability.md @@ -45,7 +45,7 @@ Each request your application makes to an LLM provider corresponds to one LLM sp | ------------------------------- | --------------------------------------------------------------------------------------------- | | OpenAI | Total requests in the OpenAI admin UI, or sum the `*.num_model_requests` metrics | | AWS Bedrock | Invocations by model in CloudWatch, or the `aws.bedrock.invocations` metric | -| Google Enterprise Agent Platform | Usage metrics in the platform console, or the `gcp.aiplatform.online_prediction_requests_per_base_model` metric | +| Google Vertex AI | Usage metrics in the Google Cloud console, or the `gcp.aiplatform.online_prediction_requests_per_base_model` metric | | Vercel | Requests by provider and model in the Vercel dashboard, or the `vercel.requests` metric | | Anthropic | Request totals are not reported; use your own application metrics as a proxy | From a65696258d7c8ce049818c7517f74829459db838 Mon Sep 17 00:00:00 2001 From: Greg Svigruha Date: Wed, 2 Sep 2026 16:22:32 -0400 Subject: [PATCH 08/23] Link pricing to the Agent Observability product section Point the pricing link at the Agent Observability product anchor, and add the billing page to the retention page's further reading. Co-Authored-By: Claude Opus 5 (1M context) --- .../content/en/account_management/billing/llm_observability.md | 2 +- hugo/content/en/llm_observability/data_retention.md | 3 +++ 2 files changed, 4 insertions(+), 1 deletion(-) diff --git a/hugo/content/en/account_management/billing/llm_observability.md b/hugo/content/en/account_management/billing/llm_observability.md index 49258a321d0..b4783943553 100644 --- a/hugo/content/en/account_management/billing/llm_observability.md +++ b/hugo/content/en/account_management/billing/llm_observability.md @@ -88,7 +88,7 @@ This allotment applies to every plan. Usage beyond your allotment is billed as S {{< partial name="whats-next/whats-next.html" >}} [1]: /llm_observability/ -[2]: https://www.datadoghq.com/pricing/ +[2]: https://www.datadoghq.com/pricing/?product=agent-observability#products [3]: /llm_observability/improve/experiments/ [4]: /llm_observability/instrument/sdk/ [5]: /llm_observability/instrument/agentic/python diff --git a/hugo/content/en/llm_observability/data_retention.md b/hugo/content/en/llm_observability/data_retention.md index 51f4a40dd39..3426ae08b1a 100644 --- a/hugo/content/en/llm_observability/data_retention.md +++ b/hugo/content/en/llm_observability/data_retention.md @@ -2,6 +2,9 @@ title: Data Retention description: Learn how long Agent Observability retains traces, experiments, annotations, datasets, prompts, and metrics. further_reading: + - link: "/account_management/billing/llm_observability/" + tag: "Documentation" + text: "Learn how Agent Observability is billed" - link: "/llm_observability/data_privacy_security_and_rbac/" tag: "Documentation" text: "Control access to sensitive Agent Observability data" From bf84424d2bb05d464cb5139e8659f674d8f99bde Mon Sep 17 00:00:00 2001 From: Greg Svigruha Date: Wed, 2 Sep 2026 16:27:14 -0400 Subject: [PATCH 09/23] Add Agent Observability to the pricing page The shared pricing page carries a section per product defining its billable unit, but had none for Agent Observability. Add one covering LLM spans, and link to the Agent Observability billing page. Co-Authored-By: Claude Opus 5 (1M context) --- hugo/content/en/account_management/billing/pricing.md | 10 ++++++++++ 1 file changed, 10 insertions(+) diff --git a/hugo/content/en/account_management/billing/pricing.md b/hugo/content/en/account_management/billing/pricing.md index a158134723d..91b2ce36eea 100644 --- a/hugo/content/en/account_management/billing/pricing.md +++ b/hugo/content/en/account_management/billing/pricing.md @@ -106,6 +106,14 @@ You can put controls in place for both Indexed and Ingested span volumes. For mo * In the event that a pipeline is not associated with a git repository, or git metadata is unavailable, the username of the person triggering the pipeline execution is used as the billable committer. * For Pipeline Visibility, every pipeline, pipeline stage, and pipeline job counts as a **pipeline span**. For Testing Visibility, every individual test run counts as a **test span**. +## Agent Observability + +* Datadog charges based on the number of **LLM spans** ingested by the Agent Observability service. An LLM span represents a single request to an LLM provider, such as OpenAI or Anthropic. +* One agent workflow can make several LLM requests, and each request produces its own LLM span. Datadog bills only on LLM spans, not on the surrounding workflow, tool, and retrieval spans in the same trace. +* [Experiments][13] are billed the same way, on the number of LLM spans ingested. +* Indexed LLM spans are included in the base price. +* For more detail, see [Agent Observability Billing][14]. + ## Troubleshooting For technical questions, contact [Datadog support][7]. @@ -124,3 +132,5 @@ Contact [Sales][8] or your [Customer Success][9] Manager to discuss hourly prici [10]: /real_user_monitoring/rum_without_limits/ [11]: https://www.datadoghq.com/pricing/?product=real-user-monitoring#products [12]: /account_management/billing/metric_name_pricing/ +[13]: /llm_observability/improve/experiments/ +[14]: /account_management/billing/llm_observability/ From 21656ae76820ee20f74ac7d6ac7c464ef7d95892 Mon Sep 17 00:00:00 2001 From: Greg Svigruha Date: Wed, 2 Sep 2026 16:34:28 -0400 Subject: [PATCH 10/23] State Agent Observability rates on the billing and pricing pages Follow the APM billing model: a rate table for the free tier and the on-demand span rates, plus worked scenarios showing how a monthly bill is calculated at three usage levels. Note that committed and enterprise pricing is negotiated separately. Co-Authored-By: Claude Opus 5 (1M context) --- .../billing/llm_observability.md | 78 ++++++++++++++----- .../en/account_management/billing/pricing.md | 3 +- 2 files changed, 59 insertions(+), 22 deletions(-) diff --git a/hugo/content/en/account_management/billing/llm_observability.md b/hugo/content/en/account_management/billing/llm_observability.md index b4783943553..056269282fe 100644 --- a/hugo/content/en/account_management/billing/llm_observability.md +++ b/hugo/content/en/account_management/billing/llm_observability.md @@ -11,19 +11,51 @@ further_reading: text: "Learn how long Agent Observability retains your data" --- -## Overview +[Agent Observability][1] is metered and billed on the number of LLM spans ingested. An LLM span represents a single request to an LLM provider, such as OpenAI or Anthropic. One agent workflow can make several LLM requests, and each request produces its own LLM span. Datadog bills only on LLM spans, not on the surrounding workflow, tool, and retrieval spans in the same trace. -This page contains common questions and answers about billing topics for [Agent Observability][1]. +| Billing Parameter | Price | Included | Billing | +|------------------------------|---------------------------------------------|-----------------------------------------------------------------------------------------------------------|----------------------------------------------------------------------------------------------------------------------------------------------------------------| +| Free tier | $0 | Up to 40,000 LLM spans per month | Organizations at or below 40,000 LLM spans in a calendar month are not billed for Agent Observability. Retention add-ons are not available on the free tier. | +| LLM spans, first 100,000 | $240 per month | All indexed LLM spans, [Experiments][3], and 1 GB of [Sensitive Data Scanner][10] usage per 10,000 LLM spans | Datadog counts the LLM spans ingested into the Agent Observability service during the calendar month. Exceeding 40,000 LLM spans moves the organization onto the paid plan. [More Agent Observability pricing information.][2] | +| Additional LLM spans | $5 per 10,000 LLM spans per month | Same as above | Billed on LLM spans beyond the first 100,000 in the calendar month, in blocks of 10,000 spans. | -## How is Agent Observability billed? +The rates above are on-demand rates. Committed monthly and annual plans offer lower rates, and enterprise pricing is custom. Contact [Sales][11] or your [Customer Success][12] Manager to discuss a commitment or volume discounts for your account. -Agent Observability is metered and billed on the number of LLM spans ingested. For current rates, see the [Datadog pricing page][2]. +## Billing scenarios -## What counts as an LLM span? +**Sample cases illustrate on-demand billing rates with the default 15-day span retention. Contact [Sales][11] or your [Customer Success][12] Manager to discuss committed pricing and volume discounts for your account.** + +### Usage within the free tier + +Sending 30,000 LLM spans in a month. + +| Billable Unit | Quantity | Price | Formula | Subtotal | +|---------------|----------|-------------------------------------------|---------|-------------------| +| LLM spans | 30,000 | First 40,000 LLM spans included at no cost | — | $0 | +| Total | 30,000 | | | **$0 per month** | + +### Usage above the free tier + +Sending 80,000 LLM spans in a month. + +| Billable Unit | Quantity | Price | Formula | Subtotal | +|---------------|----------|--------------------------------------------------|---------|---------------------| +| LLM spans | 80,000 | $240 per month for the first 100,000 LLM spans | $240 | $240 | +| Total | 80,000 | | | **$240 per month** | + +### Usage above 100,000 LLM spans -An LLM span represents a single request to an LLM provider, such as OpenAI or Anthropic. +Sending 250,000 LLM spans in a month. -One agent workflow can make several LLM requests, and each one produces its own LLM span. A trace containing seven LLM requests therefore contains seven billable LLM spans. Datadog bills only on LLM spans, not on the surrounding workflow, tool, and retrieval spans in the same trace. +| Billable Unit | Quantity | Price | Formula | Subtotal | +|--------------------------|----------|------------------------------------------------|-------------|---------------------| +| LLM spans, first 100,000 | 100,000 | $240 per month for the first 100,000 LLM spans | $240 | $240 | +| Additional LLM spans | 150,000 | $5 per 10,000 LLM spans | 15 * $5 | $75 | +| Total | 250,000 | | $240 + $75 | **$315 per month** | + +## What counts as an LLM span? + +An LLM span represents a single request to an LLM provider. A trace containing seven LLM requests therefore contains seven billable LLM spans, regardless of how many workflow, tool, or retrieval spans surround them. ## Are experiments billed separately? @@ -33,21 +65,17 @@ No. [Experiments][3] are part of Agent Observability and are billed the same way No. Indexed LLM spans are included in the base price. -## Is there a free allotment of LLM spans? - -Yes. Each organization receives an allotment of LLM spans per month at no cost. For the current allotment, see the [Datadog pricing page][2]. - ## How can I estimate my LLM span usage? Each request your application makes to an LLM provider corresponds to one LLM span, so provider-side request counts are a good starting point. Most providers report request counts in their own console, and Datadog integrations expose the same counts as metrics: -| Provider | Where to find request counts | -| ------------------------------- | --------------------------------------------------------------------------------------------- | -| OpenAI | Total requests in the OpenAI admin UI, or sum the `*.num_model_requests` metrics | -| AWS Bedrock | Invocations by model in CloudWatch, or the `aws.bedrock.invocations` metric | -| Google Vertex AI | Usage metrics in the Google Cloud console, or the `gcp.aiplatform.online_prediction_requests_per_base_model` metric | -| Vercel | Requests by provider and model in the Vercel dashboard, or the `vercel.requests` metric | -| Anthropic | Request totals are not reported; use your own application metrics as a proxy | +| Provider | Where to find request counts | +|-------------------|---------------------------------------------------------------------------------------------------------------------| +| OpenAI | Total requests in the OpenAI admin UI, or sum the `*.num_model_requests` metrics | +| AWS Bedrock | Invocations by model in CloudWatch, or the `aws.bedrock.invocations` metric | +| Google Vertex AI | Usage metrics in the Google Cloud console, or the `gcp.aiplatform.online_prediction_requests_per_base_model` metric | +| Vercel | Requests by provider and model in the Vercel dashboard, or the `vercel.requests` metric | +| Anthropic | Request totals are not reported; use your own application metrics as a proxy | Keep three things in mind when using provider request counts: @@ -63,7 +91,7 @@ In Datadog, go to [{{< ui >}}Plan & Usage{{< /ui >}}][7] and open the {{< ui >}} ## Does extending my data retention period affect billing? -Yes. Longer retention is a paid add-on, because it increases how much data Datadog stores for you. Add-ons extend span retention to 30, 60, or 90 days, along with a correspondingly longer period for experiment traces. For the periods each add-on provides, see [Data Retention][8]. To add one, contact your Datadog account representative or [Datadog support][9]. +Yes. Longer retention is a paid add-on, because it increases how much data Datadog stores for you. Add-ons extend span retention to 30, 60, or 90 days, along with a correspondingly longer period for experiment traces. Add-ons are not available on the free tier. For the periods each add-on provides, see [Data Retention][8]. To add one, contact your Datadog account representative or [Datadog support][9]. ## Is Agent Observability a standalone product? @@ -76,12 +104,18 @@ Agent Observability uses [Sensitive Data Scanner][10] Library Rules to identify Your Sensitive Data Scanner allotment scales with your LLM span usage: every 10,000 LLM spans includes 1 GB of Sensitive Data Scanner usage. | LLM spans | Included Sensitive Data Scanner usage | -| --------- | ------------------------------------- | +|-----------|---------------------------------------| | 40,000 | 4 GB | | 100,000 | 10 GB | | 300,000 | 30 GB | -This allotment applies to every plan. Usage beyond your allotment is billed as Sensitive Data Scanner usage. +This allotment applies to every plan, including the free tier. Usage beyond your allotment is billed as Sensitive Data Scanner usage. + +## Troubleshooting + +For technical questions, contact [Datadog support][9]. + +Contact [Sales][11] or your [Customer Success][12] Manager to discuss pricing or billing for your account. ## Further reading @@ -97,3 +131,5 @@ This allotment applies to every plan. Usage beyond your allotment is billed as S [8]: /llm_observability/data_retention/ [9]: /help/ [10]: /security/sensitive_data_scanner/ +[11]: mailto:sales@datadoghq.com +[12]: mailto:success@datadoghq.com diff --git a/hugo/content/en/account_management/billing/pricing.md b/hugo/content/en/account_management/billing/pricing.md index 91b2ce36eea..3eeab02876c 100644 --- a/hugo/content/en/account_management/billing/pricing.md +++ b/hugo/content/en/account_management/billing/pricing.md @@ -112,7 +112,8 @@ You can put controls in place for both Indexed and Ingested span volumes. For mo * One agent workflow can make several LLM requests, and each request produces its own LLM span. Datadog bills only on LLM spans, not on the surrounding workflow, tool, and retrieval spans in the same trace. * [Experiments][13] are billed the same way, on the number of LLM spans ingested. * Indexed LLM spans are included in the base price. -* For more detail, see [Agent Observability Billing][14]. +* On on-demand plans, the first 40,000 LLM spans each month are free. Above that, Datadog charges $240 per month for the first 100,000 LLM spans, then $5 per additional 10,000 LLM spans. Committed and enterprise pricing is lower and negotiated. +* For rate tables and worked examples, see [Agent Observability Billing][14]. ## Troubleshooting From 7c5ad8061859591b1130742712150d9e6ef28daf Mon Sep 17 00:00:00 2001 From: Greg Svigruha Date: Wed, 2 Sep 2026 16:40:20 -0400 Subject: [PATCH 11/23] Keep rates off the shared pricing page Sections on the pricing page define each product's billable unit and leave rates to the product billing page, as the APM section does. Drop the duplicated Agent Observability rates so they have a single owner. Co-Authored-By: Claude Opus 5 (1M context) --- hugo/content/en/account_management/billing/pricing.md | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/hugo/content/en/account_management/billing/pricing.md b/hugo/content/en/account_management/billing/pricing.md index 3eeab02876c..e532ab60c00 100644 --- a/hugo/content/en/account_management/billing/pricing.md +++ b/hugo/content/en/account_management/billing/pricing.md @@ -112,8 +112,8 @@ You can put controls in place for both Indexed and Ingested span volumes. For mo * One agent workflow can make several LLM requests, and each request produces its own LLM span. Datadog bills only on LLM spans, not on the surrounding workflow, tool, and retrieval spans in the same trace. * [Experiments][13] are billed the same way, on the number of LLM spans ingested. * Indexed LLM spans are included in the base price. -* On on-demand plans, the first 40,000 LLM spans each month are free. Above that, Datadog charges $240 per month for the first 100,000 LLM spans, then $5 per additional 10,000 LLM spans. Committed and enterprise pricing is lower and negotiated. -* For rate tables and worked examples, see [Agent Observability Billing][14]. +* On-demand plans include an allotment of LLM spans each month at no cost, and are billed in blocks of LLM spans beyond it. Committed and enterprise pricing is negotiated separately. +* For rates, worked examples, and usage estimation, see [Agent Observability Billing][14]. ## Troubleshooting From a661ef854f09bac664795f3e059c8f9f81aac62d Mon Sep 17 00:00:00 2001 From: Greg Svigruha Date: Wed, 2 Sep 2026 16:48:13 -0400 Subject: [PATCH 12/23] Confirm annotation retention as 90 days at no charge State the annotation label period as 90 days rather than deferring to the annotated object, and note that extending retention by annotating is not billed. Co-Authored-By: Claude Opus 5 (1M context) --- .../en/account_management/billing/llm_observability.md | 2 ++ hugo/content/en/llm_observability/data_retention.md | 5 ++++- .../en/llm_observability/investigate/annotation_queues.md | 8 ++++---- 3 files changed, 10 insertions(+), 5 deletions(-) diff --git a/hugo/content/en/account_management/billing/llm_observability.md b/hugo/content/en/account_management/billing/llm_observability.md index 056269282fe..36f221c85f9 100644 --- a/hugo/content/en/account_management/billing/llm_observability.md +++ b/hugo/content/en/account_management/billing/llm_observability.md @@ -93,6 +93,8 @@ In Datadog, go to [{{< ui >}}Plan & Usage{{< /ui >}}][7] and open the {{< ui >}} Yes. Longer retention is a paid add-on, because it increases how much data Datadog stores for you. Add-ons extend span retention to 30, 60, or 90 days, along with a correspondingly longer period for experiment traces. Add-ons are not available on the free tier. For the periods each add-on provides, see [Data Retention][8]. To add one, contact your Datadog account representative or [Datadog support][9]. +Annotating a trace, span, or session also extends its retention to 90 days, and that extension is not billed. + ## Is Agent Observability a standalone product? Yes. Agent Observability does not require you to purchase any other Datadog product. diff --git a/hugo/content/en/llm_observability/data_retention.md b/hugo/content/en/llm_observability/data_retention.md index 3426ae08b1a..3bbb91ad560 100644 --- a/hugo/content/en/llm_observability/data_retention.md +++ b/hugo/content/en/llm_observability/data_retention.md @@ -26,6 +26,7 @@ Retention periods in Agent Observability depend on the type of data and on your | Experiment traces | 15 days on on-demand plans; 90 days on committed plans; 6, 9, or 12 months with a retention add-on | | Experiment definitions and aggregate results | 90 days from creation | | Annotated traces, spans, and sessions | 90 days from the time of annotation, or your span retention period if that is longer | +| Annotation labels | 90 days, matching the object they annotate | | Dataset records | 3 years, regardless of your span retention period | | Prompts in the prompt registry | 3 years, extended each time the prompt is pulled | | `ml_obs.*` metrics | 15 months | @@ -70,7 +71,9 @@ Annotating an object extends its retention. When you apply an annotation label o If your organization's span retention period is longer than 90 days, annotated objects are retained for that longer period instead. -Annotation labels are viewable for as long as the object they annotate is retained. +Annotation labels are retained for the same 90 days as the object they annotate, and are no longer viewable after it expires. + +Extending retention by annotating an object does not incur an additional charge. Free-form notes are not attached to a trace, span, or session, so no object's retention applies to them. diff --git a/hugo/content/en/llm_observability/investigate/annotation_queues.md b/hugo/content/en/llm_observability/investigate/annotation_queues.md index 4202052c531..51fe4024e2f 100644 --- a/hugo/content/en/llm_observability/investigate/annotation_queues.md +++ b/hugo/content/en/llm_observability/investigate/annotation_queues.md @@ -325,11 +325,11 @@ You can manage annotation queues programmatically. The following endpoints are a | Data | Retention period | | ----------------- | ----------------------------------------------------| -| Traces in queues | Not retained beyond your organization's trace retention period, unless annotated | -| Annotated traces | 90 days from the time of annotation, or your trace retention period if that is longer | -| Annotation labels | As long as the trace they annotate | +| Traces in queues | Not retained beyond your organization's span retention period, unless annotated | +| Annotated traces | 90 days from the time of annotation, or your span retention period if that is longer | +| Annotation labels | 90 days, matching the trace they annotate | -Annotating a trace extends its retention: a trace that would otherwise expire under a shorter trace retention period is retained for 90 days from the time you annotate it. For details, see [Data Retention][16]. +Annotating a trace extends its retention at no additional charge: a trace that would otherwise expire under a shorter span retention period is retained for 90 days from the time you annotate it. For details, see [Data Retention][16]. ## Example workflows From 67278f13a39546fb120d53b04dd40f113087b439 Mon Sep 17 00:00:00 2001 From: Greg Svigruha Date: Wed, 2 Sep 2026 16:49:32 -0400 Subject: [PATCH 13/23] Keep annotation retention on the retention page only Co-Authored-By: Claude Opus 5 (1M context) --- hugo/content/en/account_management/billing/llm_observability.md | 2 -- 1 file changed, 2 deletions(-) diff --git a/hugo/content/en/account_management/billing/llm_observability.md b/hugo/content/en/account_management/billing/llm_observability.md index 36f221c85f9..056269282fe 100644 --- a/hugo/content/en/account_management/billing/llm_observability.md +++ b/hugo/content/en/account_management/billing/llm_observability.md @@ -93,8 +93,6 @@ In Datadog, go to [{{< ui >}}Plan & Usage{{< /ui >}}][7] and open the {{< ui >}} Yes. Longer retention is a paid add-on, because it increases how much data Datadog stores for you. Add-ons extend span retention to 30, 60, or 90 days, along with a correspondingly longer period for experiment traces. Add-ons are not available on the free tier. For the periods each add-on provides, see [Data Retention][8]. To add one, contact your Datadog account representative or [Datadog support][9]. -Annotating a trace, span, or session also extends its retention to 90 days, and that extension is not billed. - ## Is Agent Observability a standalone product? Yes. Agent Observability does not require you to purchase any other Datadog product. From dad9e8132b079e3257de67b9e262ca08743391f7 Mon Sep 17 00:00:00 2001 From: Greg Svigruha Date: Wed, 2 Sep 2026 16:52:52 -0400 Subject: [PATCH 14/23] Document the capabilities included in the base price Replace the separate questions about experiments and indexed spans with one section listing everything included and billed only on LLM spans: evaluations, experiments, patterns, annotation queues, datasets, prompt management, and the playground. Tie the Sensitive Data Scanner section to that list, since it is the one inclusion with a usage allotment. Co-Authored-By: Claude Opus 5 (1M context) --- .../billing/llm_observability.md | 25 ++++++++++++++----- 1 file changed, 19 insertions(+), 6 deletions(-) diff --git a/hugo/content/en/account_management/billing/llm_observability.md b/hugo/content/en/account_management/billing/llm_observability.md index 056269282fe..6b4a9521fe6 100644 --- a/hugo/content/en/account_management/billing/llm_observability.md +++ b/hugo/content/en/account_management/billing/llm_observability.md @@ -57,13 +57,18 @@ Sending 250,000 LLM spans in a month. An LLM span represents a single request to an LLM provider. A trace containing seven LLM requests therefore contains seven billable LLM spans, regardless of how many workflow, tool, or retrieval spans surround them. -## Are experiments billed separately? +## What is included in the base price? -No. [Experiments][3] are part of Agent Observability and are billed the same way as production monitoring, on the number of LLM spans ingested. +Every Agent Observability capability is included in the base price and billed only on the LLM spans ingested. There is no separate charge, quota, or add-on for: -## Are indexed spans billed separately? - -No. Indexed LLM spans are included in the base price. +- Indexed LLM spans +- [Evaluations][13], including managed evaluations and custom LLM-as-a-judge evaluations, with no limit on how many you run +- [Experiments][3], which are metered on LLM spans in the same way as production monitoring +- [Patterns][14] +- [Annotation queues][15] +- [Datasets][16] +- [Prompt management][17] and the [playground][18] +- [Sensitive Data Scanner](#is-sensitive-data-scanner-included), within the allotment described below ## How can I estimate my LLM span usage? @@ -99,7 +104,9 @@ Yes. Agent Observability does not require you to purchase any other Datadog prod ## Is Sensitive Data Scanner included? -Agent Observability uses [Sensitive Data Scanner][10] Library Rules to identify and redact sensitive information in your LLM application traffic, including personal information, financial data, and health records. You do not purchase Sensitive Data Scanner separately to use this capability. +Yes. As with the other capabilities [included in the base price](#what-is-included-in-the-base-price), you do not purchase Sensitive Data Scanner separately. Agent Observability uses [Sensitive Data Scanner][10] Library Rules to identify and redact sensitive information in your LLM application traffic, including personal information, financial data, and health records. + +Sensitive Data Scanner differs from the others in having a usage allotment rather than being unmetered. Your Sensitive Data Scanner allotment scales with your LLM span usage: every 10,000 LLM spans includes 1 GB of Sensitive Data Scanner usage. @@ -133,3 +140,9 @@ Contact [Sales][11] or your [Customer Success][12] Manager to discuss pricing or [10]: /security/sensitive_data_scanner/ [11]: mailto:sales@datadoghq.com [12]: mailto:success@datadoghq.com +[13]: /llm_observability/investigate/evaluations/ +[14]: /llm_observability/investigate/patterns/ +[15]: /llm_observability/investigate/annotation_queues/ +[16]: /llm_observability/improve/datasets/ +[17]: /llm_observability/configure/prompt_management/ +[18]: /llm_observability/improve/playground From 8d0a917ceb7005dd920b6d060e426f81323d3955 Mon Sep 17 00:00:00 2001 From: Greg Svigruha Date: Wed, 2 Sep 2026 16:54:41 -0400 Subject: [PATCH 15/23] Say how to change a retention period, in one place Move the how-to into the Changing your retention period section, where a reader looks for it, and have the traces section link down to it rather than repeating the contact route. Co-Authored-By: Claude Opus 5 (1M context) --- hugo/content/en/llm_observability/data_retention.md | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/hugo/content/en/llm_observability/data_retention.md b/hugo/content/en/llm_observability/data_retention.md index 3bbb91ad560..5fc6661f0ea 100644 --- a/hugo/content/en/llm_observability/data_retention.md +++ b/hugo/content/en/llm_observability/data_retention.md @@ -35,7 +35,7 @@ Retention periods in Agent Observability depend on the type of data and on your Traces and spans from your instrumented applications are retained for **15 days** on all plans by default. This applies to everything stored on the span, including per-span operational data such as cost, token counts, latency, and errors, as well as evaluation scores attached to spans. -A retention add-on extends this to **30, 60, or 90 days**. To request a longer retention period, contact your Datadog account representative or [Datadog support][1]. For how retention affects billing, see [Agent Observability Billing][10]. +A retention add-on extends this to **30, 60, or 90 days**. See [Changing your retention period](#changing-your-retention-period). Retention applies to the raw spans you query in the Traces explorer. Metrics derived from those spans are retained separately, for longer. See [Metrics](#metrics). @@ -59,6 +59,8 @@ The experiment itself — its name, configuration, and aggregate results — is ## Changing your retention period +Retention add-ons are arranged through your account team rather than enabled from the Datadog UI. To request a longer retention period, contact your Datadog account representative or [Datadog support][1]. For how add-ons are priced, see [Agent Observability Billing][10]. + When you add or extend a retention add-on, the longer period applies **retroactively to every span that has not already expired**. Spans that expired under your previous period are not recoverable. For example, if you are on the default 15-day retention and add a 60-day add-on today, the spans from the last 15 days pick up the 60-day period, but anything older is already gone. From fa7a12bf61c16b1da2d8fe7549a8710c315d35df Mon Sep 17 00:00:00 2001 From: Greg Svigruha Date: Wed, 2 Sep 2026 16:55:18 -0400 Subject: [PATCH 16/23] Simplify the included-features section State that all features are included and nothing is charged extra, rather than enumerating them, and keep the Sensitive Data Scanner allotment as the stated exception. Co-Authored-By: Claude Opus 5 (1M context) --- .../billing/llm_observability.md | 17 ++--------------- 1 file changed, 2 insertions(+), 15 deletions(-) diff --git a/hugo/content/en/account_management/billing/llm_observability.md b/hugo/content/en/account_management/billing/llm_observability.md index 6b4a9521fe6..b88f96f63fe 100644 --- a/hugo/content/en/account_management/billing/llm_observability.md +++ b/hugo/content/en/account_management/billing/llm_observability.md @@ -59,16 +59,9 @@ An LLM span represents a single request to an LLM provider. A trace containing s ## What is included in the base price? -Every Agent Observability capability is included in the base price and billed only on the LLM spans ingested. There is no separate charge, quota, or add-on for: +All Agent Observability features are included in the base price. Datadog charges only for the LLM spans you ingest, and does not charge extra for any capability. -- Indexed LLM spans -- [Evaluations][13], including managed evaluations and custom LLM-as-a-judge evaluations, with no limit on how many you run -- [Experiments][3], which are metered on LLM spans in the same way as production monitoring -- [Patterns][14] -- [Annotation queues][15] -- [Datasets][16] -- [Prompt management][17] and the [playground][18] -- [Sensitive Data Scanner](#is-sensitive-data-scanner-included), within the allotment described below +The one exception is [Sensitive Data Scanner](#is-sensitive-data-scanner-included), which is included within a usage allotment that scales with your LLM span usage. ## How can I estimate my LLM span usage? @@ -140,9 +133,3 @@ Contact [Sales][11] or your [Customer Success][12] Manager to discuss pricing or [10]: /security/sensitive_data_scanner/ [11]: mailto:sales@datadoghq.com [12]: mailto:success@datadoghq.com -[13]: /llm_observability/investigate/evaluations/ -[14]: /llm_observability/investigate/patterns/ -[15]: /llm_observability/investigate/annotation_queues/ -[16]: /llm_observability/improve/datasets/ -[17]: /llm_observability/configure/prompt_management/ -[18]: /llm_observability/improve/playground From 210e2d5d39d65e6483eea271ca6cc0c29367f0a5 Mon Sep 17 00:00:00 2001 From: Greg Svigruha Date: Wed, 2 Sep 2026 16:56:13 -0400 Subject: [PATCH 17/23] Note the BYOK option for evaluations and patterns Evaluations and patterns can run with your own LLM provider key instead of the Datadog-provided model, in which case your provider bills you for that model usage directly. Co-Authored-By: Claude Opus 5 (1M context) --- .../en/account_management/billing/llm_observability.md | 5 +++++ 1 file changed, 5 insertions(+) diff --git a/hugo/content/en/account_management/billing/llm_observability.md b/hugo/content/en/account_management/billing/llm_observability.md index b88f96f63fe..56e8ff82f13 100644 --- a/hugo/content/en/account_management/billing/llm_observability.md +++ b/hugo/content/en/account_management/billing/llm_observability.md @@ -63,6 +63,8 @@ All Agent Observability features are included in the base price. Datadog charges The one exception is [Sensitive Data Scanner](#is-sensitive-data-scanner-included), which is included within a usage allotment that scales with your LLM span usage. +[Evaluations][13] and [patterns][14] use an LLM to do their work. You can optionally run them with your own LLM provider key (bring your own key, or BYOK) instead of the model Datadog provides. Your LLM provider bills you directly for the model usage in that case, separately from your Datadog bill. To connect a provider, see [Connect your LLM provider account][15]. + ## How can I estimate my LLM span usage? Each request your application makes to an LLM provider corresponds to one LLM span, so provider-side request counts are a good starting point. Most providers report request counts in their own console, and Datadog integrations expose the same counts as metrics: @@ -133,3 +135,6 @@ Contact [Sales][11] or your [Customer Success][12] Manager to discuss pricing or [10]: /security/sensitive_data_scanner/ [11]: mailto:sales@datadoghq.com [12]: mailto:success@datadoghq.com +[13]: /llm_observability/investigate/evaluations/ +[14]: /llm_observability/investigate/patterns/ +[15]: /llm_observability/investigate/evaluations/llm_as_a_judge_evaluations/connect_to_account/ From 24b5560ccc75da1cf23e00689d8941445c6388cf Mon Sep 17 00:00:00 2001 From: Greg Svigruha Date: Wed, 2 Sep 2026 16:57:13 -0400 Subject: [PATCH 18/23] Make BYOK the condition for no extra charge on evals and patterns Evaluations and patterns are not charged extra when run with the customer's own LLM provider key, rather than BYOK being an alternative to an unmetered default. State annotations and datasets as included, and experiments as metered on LLM spans like production traces. Co-Authored-By: Claude Opus 5 (1M context) --- .../en/account_management/billing/llm_observability.md | 10 ++++++---- 1 file changed, 6 insertions(+), 4 deletions(-) diff --git a/hugo/content/en/account_management/billing/llm_observability.md b/hugo/content/en/account_management/billing/llm_observability.md index 56e8ff82f13..96735ac2653 100644 --- a/hugo/content/en/account_management/billing/llm_observability.md +++ b/hugo/content/en/account_management/billing/llm_observability.md @@ -59,11 +59,11 @@ An LLM span represents a single request to an LLM provider. A trace containing s ## What is included in the base price? -All Agent Observability features are included in the base price. Datadog charges only for the LLM spans you ingest, and does not charge extra for any capability. +Datadog charges only for the LLM spans you ingest. [Annotation queues][15] and [datasets][16] are included at no extra charge, and [experiments][3] are metered on LLM spans in the same way as production traces. -The one exception is [Sensitive Data Scanner](#is-sensitive-data-scanner-included), which is included within a usage allotment that scales with your LLM span usage. +[Evaluations][13] and [patterns][14] use an LLM to do their work. Datadog does not charge extra for them when you run them with your own LLM provider key (bring your own key, or BYOK). Your LLM provider bills you directly for that model usage, separately from your Datadog bill. To connect a provider, see [Connect your LLM provider account][17]. -[Evaluations][13] and [patterns][14] use an LLM to do their work. You can optionally run them with your own LLM provider key (bring your own key, or BYOK) instead of the model Datadog provides. Your LLM provider bills you directly for the model usage in that case, separately from your Datadog bill. To connect a provider, see [Connect your LLM provider account][15]. +[Sensitive Data Scanner](#is-sensitive-data-scanner-included) is included within a usage allotment that scales with your LLM span usage. ## How can I estimate my LLM span usage? @@ -137,4 +137,6 @@ Contact [Sales][11] or your [Customer Success][12] Manager to discuss pricing or [12]: mailto:success@datadoghq.com [13]: /llm_observability/investigate/evaluations/ [14]: /llm_observability/investigate/patterns/ -[15]: /llm_observability/investigate/evaluations/llm_as_a_judge_evaluations/connect_to_account/ +[15]: /llm_observability/investigate/annotation_queues/ +[16]: /llm_observability/improve/datasets/ +[17]: /llm_observability/investigate/evaluations/llm_as_a_judge_evaluations/connect_to_account/ From 988f1f3cb0a914dc892076b378b5ea8f2aa29d45 Mon Sep 17 00:00:00 2001 From: Greg Svigruha Date: Wed, 2 Sep 2026 16:57:38 -0400 Subject: [PATCH 19/23] Note the free preview for Datadog-provided evaluation models Evaluations can run without a customer provider key through a free preview with limited capacity, so BYOK is the route for production volume. Co-Authored-By: Claude Opus 5 (1M context) --- hugo/content/en/account_management/billing/llm_observability.md | 2 ++ 1 file changed, 2 insertions(+) diff --git a/hugo/content/en/account_management/billing/llm_observability.md b/hugo/content/en/account_management/billing/llm_observability.md index 96735ac2653..0da508874ec 100644 --- a/hugo/content/en/account_management/billing/llm_observability.md +++ b/hugo/content/en/account_management/billing/llm_observability.md @@ -63,6 +63,8 @@ Datadog charges only for the LLM spans you ingest. [Annotation queues][15] and [ [Evaluations][13] and [patterns][14] use an LLM to do their work. Datadog does not charge extra for them when you run them with your own LLM provider key (bring your own key, or BYOK). Your LLM provider bills you directly for that model usage, separately from your Datadog bill. To connect a provider, see [Connect your LLM provider account][17]. +Evaluations can also run on a model Datadog provides, without a provider key of your own. This option is a free preview with limited capacity, so bringing your own key is the route for running evaluations at production volume. + [Sensitive Data Scanner](#is-sensitive-data-scanner-included) is included within a usage allotment that scales with your LLM span usage. ## How can I estimate my LLM span usage? From d7210b69ed903b8296c1268e4f276c982cc2b386 Mon Sep 17 00:00:00 2001 From: Greg Svigruha Date: Wed, 2 Sep 2026 16:58:05 -0400 Subject: [PATCH 20/23] Present BYOK as an option alongside the free default Evaluations and patterns are not charged extra either way: they run on a Datadog-provided model by default, with BYOK as an option for production volume. Co-Authored-By: Claude Opus 5 (1M context) --- .../en/account_management/billing/llm_observability.md | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/hugo/content/en/account_management/billing/llm_observability.md b/hugo/content/en/account_management/billing/llm_observability.md index 0da508874ec..c3206cbc5c1 100644 --- a/hugo/content/en/account_management/billing/llm_observability.md +++ b/hugo/content/en/account_management/billing/llm_observability.md @@ -61,9 +61,9 @@ An LLM span represents a single request to an LLM provider. A trace containing s Datadog charges only for the LLM spans you ingest. [Annotation queues][15] and [datasets][16] are included at no extra charge, and [experiments][3] are metered on LLM spans in the same way as production traces. -[Evaluations][13] and [patterns][14] use an LLM to do their work. Datadog does not charge extra for them when you run them with your own LLM provider key (bring your own key, or BYOK). Your LLM provider bills you directly for that model usage, separately from your Datadog bill. To connect a provider, see [Connect your LLM provider account][17]. +[Evaluations][13] and [patterns][14] use an LLM to do their work, and Datadog does not charge extra for either of them. By default they run on a model Datadog provides, which is a free preview with limited capacity. -Evaluations can also run on a model Datadog provides, without a provider key of your own. This option is a free preview with limited capacity, so bringing your own key is the route for running evaluations at production volume. +You can optionally run them with your own LLM provider key (bring your own key, or BYOK) instead, which is the route for running at production volume. Your LLM provider bills you directly for that model usage, separately from your Datadog bill. To connect a provider, see [Connect your LLM provider account][17]. [Sensitive Data Scanner](#is-sensitive-data-scanner-included) is included within a usage allotment that scales with your LLM span usage. From 046226ade662d4816ca96d06e044122461918605 Mon Sep 17 00:00:00 2001 From: Greg Svigruha Date: Wed, 2 Sep 2026 17:00:01 -0400 Subject: [PATCH 21/23] Condense the included-features section State that all features are included rather than listing them, and reduce the evaluations and patterns note to the two model options. Co-Authored-By: Claude Opus 5 (1M context) --- .../en/account_management/billing/llm_observability.md | 10 +++------- 1 file changed, 3 insertions(+), 7 deletions(-) diff --git a/hugo/content/en/account_management/billing/llm_observability.md b/hugo/content/en/account_management/billing/llm_observability.md index c3206cbc5c1..84cde59ecb2 100644 --- a/hugo/content/en/account_management/billing/llm_observability.md +++ b/hugo/content/en/account_management/billing/llm_observability.md @@ -59,11 +59,9 @@ An LLM span represents a single request to an LLM provider. A trace containing s ## What is included in the base price? -Datadog charges only for the LLM spans you ingest. [Annotation queues][15] and [datasets][16] are included at no extra charge, and [experiments][3] are metered on LLM spans in the same way as production traces. +All Agent Observability features are included in the base price. Datadog charges only for the LLM spans you ingest. -[Evaluations][13] and [patterns][14] use an LLM to do their work, and Datadog does not charge extra for either of them. By default they run on a model Datadog provides, which is a free preview with limited capacity. - -You can optionally run them with your own LLM provider key (bring your own key, or BYOK) instead, which is the route for running at production volume. Your LLM provider bills you directly for that model usage, separately from your Datadog bill. To connect a provider, see [Connect your LLM provider account][17]. +[Evaluations][13] and [patterns][14] use an LLM to do their work, and Datadog does not charge extra for it. Either connect your own LLM provider account, or use the limited free preview of a Datadog-provided model. To connect a provider, see [Connect your LLM provider account][15]. [Sensitive Data Scanner](#is-sensitive-data-scanner-included) is included within a usage allotment that scales with your LLM span usage. @@ -139,6 +137,4 @@ Contact [Sales][11] or your [Customer Success][12] Manager to discuss pricing or [12]: mailto:success@datadoghq.com [13]: /llm_observability/investigate/evaluations/ [14]: /llm_observability/investigate/patterns/ -[15]: /llm_observability/investigate/annotation_queues/ -[16]: /llm_observability/improve/datasets/ -[17]: /llm_observability/investigate/evaluations/llm_as_a_judge_evaluations/connect_to_account/ +[15]: /llm_observability/investigate/evaluations/llm_as_a_judge_evaluations/connect_to_account/ From fe0d5248484d2570381ebc6ca574f3ae2c0985db Mon Sep 17 00:00:00 2001 From: Greg Svigruha Date: Wed, 2 Sep 2026 17:02:37 -0400 Subject: [PATCH 22/23] Reduce the evaluations and patterns note to the BYOK option Co-Authored-By: Claude Opus 5 (1M context) --- hugo/content/en/account_management/billing/llm_observability.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/hugo/content/en/account_management/billing/llm_observability.md b/hugo/content/en/account_management/billing/llm_observability.md index 84cde59ecb2..4200b4cde4d 100644 --- a/hugo/content/en/account_management/billing/llm_observability.md +++ b/hugo/content/en/account_management/billing/llm_observability.md @@ -61,7 +61,7 @@ An LLM span represents a single request to an LLM provider. A trace containing s All Agent Observability features are included in the base price. Datadog charges only for the LLM spans you ingest. -[Evaluations][13] and [patterns][14] use an LLM to do their work, and Datadog does not charge extra for it. Either connect your own LLM provider account, or use the limited free preview of a Datadog-provided model. To connect a provider, see [Connect your LLM provider account][15]. +You can connect your own [LLM provider account][15] for [Evaluations][13] and [Patterns][14]. [Sensitive Data Scanner](#is-sensitive-data-scanner-included) is included within a usage allotment that scales with your LLM span usage. From 5bf8d1d027874f20a18e02c8adc8ca7af45dd00c Mon Sep 17 00:00:00 2001 From: Greg Svigruha Date: Wed, 2 Sep 2026 17:07:06 -0400 Subject: [PATCH 23/23] Review pass: links, grammar, and conventions Links: point the metric retention links at the canonical Data Retention Periods page rather than one of its aliases, and add that page to further reading. Grammar: fix a dangling qualifier in the experiment traces intro, an ambiguous pronoun in the annotation labels sentence, and the repeated preposition in the experiment traces table. Say that a free-form note does not extend retention, rather than that no retention applies to it. Conventions: use Trace Explorer, the name used elsewhere in these docs. Deduplication: the billing page defined an LLM span twice, carried the sales contact three times, and stated the Sensitive Data Scanner allotment in three places. Each now appears once. Co-Authored-By: Claude Opus 5 (1M context) --- .../billing/llm_observability.md | 12 +++++------- .../en/llm_observability/data_retention.md | 15 +++++++++------ .../en/llm_observability/investigate/metrics.md | 2 +- 3 files changed, 15 insertions(+), 14 deletions(-) diff --git a/hugo/content/en/account_management/billing/llm_observability.md b/hugo/content/en/account_management/billing/llm_observability.md index 4200b4cde4d..a854c62f8b9 100644 --- a/hugo/content/en/account_management/billing/llm_observability.md +++ b/hugo/content/en/account_management/billing/llm_observability.md @@ -11,7 +11,7 @@ further_reading: text: "Learn how long Agent Observability retains your data" --- -[Agent Observability][1] is metered and billed on the number of LLM spans ingested. An LLM span represents a single request to an LLM provider, such as OpenAI or Anthropic. One agent workflow can make several LLM requests, and each request produces its own LLM span. Datadog bills only on LLM spans, not on the surrounding workflow, tool, and retrieval spans in the same trace. +[Agent Observability][1] is metered and billed on the number of LLM spans ingested, at the rates below. | Billing Parameter | Price | Included | Billing | |------------------------------|---------------------------------------------|-----------------------------------------------------------------------------------------------------------|----------------------------------------------------------------------------------------------------------------------------------------------------------------| @@ -19,7 +19,7 @@ further_reading: | LLM spans, first 100,000 | $240 per month | All indexed LLM spans, [Experiments][3], and 1 GB of [Sensitive Data Scanner][10] usage per 10,000 LLM spans | Datadog counts the LLM spans ingested into the Agent Observability service during the calendar month. Exceeding 40,000 LLM spans moves the organization onto the paid plan. [More Agent Observability pricing information.][2] | | Additional LLM spans | $5 per 10,000 LLM spans per month | Same as above | Billed on LLM spans beyond the first 100,000 in the calendar month, in blocks of 10,000 spans. | -The rates above are on-demand rates. Committed monthly and annual plans offer lower rates, and enterprise pricing is custom. Contact [Sales][11] or your [Customer Success][12] Manager to discuss a commitment or volume discounts for your account. +The rates above are on-demand rates. Committed monthly and annual plans offer lower rates, and enterprise pricing is custom. ## Billing scenarios @@ -63,7 +63,7 @@ All Agent Observability features are included in the base price. Datadog charges You can connect your own [LLM provider account][15] for [Evaluations][13] and [Patterns][14]. -[Sensitive Data Scanner](#is-sensitive-data-scanner-included) is included within a usage allotment that scales with your LLM span usage. +[Sensitive Data Scanner](#is-sensitive-data-scanner-included) is included within a usage allotment. ## How can I estimate my LLM span usage? @@ -77,7 +77,7 @@ Each request your application makes to an LLM provider corresponds to one LLM sp | Vercel | Requests by provider and model in the Vercel dashboard, or the `vercel.requests` metric | | Anthropic | Request totals are not reported; use your own application metrics as a proxy | -Keep three things in mind when using provider request counts: +Keep the following in mind when using provider request counts: - They are closer to an upper bound than an exact figure, because not every request is instrumented as an LLM span. - If you [sample][4] your traces, your billable spans are a fraction of your total requests. @@ -91,7 +91,7 @@ In Datadog, go to [{{< ui >}}Plan & Usage{{< /ui >}}][7] and open the {{< ui >}} ## Does extending my data retention period affect billing? -Yes. Longer retention is a paid add-on, because it increases how much data Datadog stores for you. Add-ons extend span retention to 30, 60, or 90 days, along with a correspondingly longer period for experiment traces. Add-ons are not available on the free tier. For the periods each add-on provides, see [Data Retention][8]. To add one, contact your Datadog account representative or [Datadog support][9]. +Yes. Longer retention is a paid add-on, because it increases how much data Datadog stores for you. Add-ons extend span retention to 30, 60, or 90 days, along with a correspondingly longer period for experiment traces. Add-ons are not available on the free tier. For the periods each add-on provides and how to request one, see [Data Retention][8]. ## Is Agent Observability a standalone product? @@ -101,8 +101,6 @@ Yes. Agent Observability does not require you to purchase any other Datadog prod Yes. As with the other capabilities [included in the base price](#what-is-included-in-the-base-price), you do not purchase Sensitive Data Scanner separately. Agent Observability uses [Sensitive Data Scanner][10] Library Rules to identify and redact sensitive information in your LLM application traffic, including personal information, financial data, and health records. -Sensitive Data Scanner differs from the others in having a usage allotment rather than being unmetered. - Your Sensitive Data Scanner allotment scales with your LLM span usage: every 10,000 LLM spans includes 1 GB of Sensitive Data Scanner usage. | LLM spans | Included Sensitive Data Scanner usage | diff --git a/hugo/content/en/llm_observability/data_retention.md b/hugo/content/en/llm_observability/data_retention.md index 5fc6661f0ea..879bc361a79 100644 --- a/hugo/content/en/llm_observability/data_retention.md +++ b/hugo/content/en/llm_observability/data_retention.md @@ -14,6 +14,9 @@ further_reading: - link: "/llm_observability/investigate/annotation_queues/" tag: "Documentation" text: "Review traces with annotation queues" + - link: "/data_security/data_retention_periods/" + tag: "Documentation" + text: "See default data retention periods across Datadog products" --- Retention periods in Agent Observability depend on the type of data and on your plan. Traces from your instrumented applications follow the span retention period in your plan, while experiment definitions, datasets, and prompts have their own periods. @@ -23,7 +26,7 @@ Retention periods in Agent Observability depend on the type of data and on your | Data | Retention period | | -------------------------------------------- | ----------------------------------------------------------------------------------------- | | Traces and spans | 15 days; 30, 60, or 90 days with a retention add-on | -| Experiment traces | 15 days on on-demand plans; 90 days on committed plans; 6, 9, or 12 months with a retention add-on | +| Experiment traces | On-demand plans: 15 days. Committed plans: 90 days. With a retention add-on: 6, 9, or 12 months | | Experiment definitions and aggregate results | 90 days from creation | | Annotated traces, spans, and sessions | 90 days from the time of annotation, or your span retention period if that is longer | | Annotation labels | 90 days, matching the object they annotate | @@ -37,11 +40,11 @@ Traces and spans from your instrumented applications are retained for **15 days* A retention add-on extends this to **30, 60, or 90 days**. See [Changing your retention period](#changing-your-retention-period). -Retention applies to the raw spans you query in the Traces explorer. Metrics derived from those spans are retained separately, for longer. See [Metrics](#metrics). +Retention applies to the raw spans you query in the Trace Explorer. Metrics derived from those spans are retained separately, for longer. See [Metrics](#metrics). ## Experiment traces -Traces produced by [experiment][3] runs are retained longer than production traces on committed plans. +On committed plans, the traces produced by [experiment][3] runs are retained longer than production traces. | Plan | Experiment trace retention | | ----------------------------------- | -------------------------- | @@ -73,11 +76,11 @@ Annotating an object extends its retention. When you apply an annotation label o If your organization's span retention period is longer than 90 days, annotated objects are retained for that longer period instead. -Annotation labels are retained for the same 90 days as the object they annotate, and are no longer viewable after it expires. +Annotation labels are retained for the same 90 days as the object they annotate, and are no longer viewable after that object expires. Extending retention by annotating an object does not incur an additional charge. -Free-form notes are not attached to a trace, span, or session, so no object's retention applies to them. +A free-form note is not attached to a trace, span, or session, so adding one does not extend any object's retention. ## Dataset records @@ -105,7 +108,7 @@ For the full list of available metrics, see [Agent Observability metrics][8]. [4]: /llm_observability/improve/datasets/ [5]: /llm_observability/improve/datasets/#dataset-versioning [6]: /metrics/ -[7]: /developers/guide/data-collection-resolution-retention/ +[7]: /data_security/data_retention_periods/ [8]: /llm_observability/investigate/metrics/ [9]: /llm_observability/configure/prompt_management/ [10]: /account_management/billing/llm_observability/ diff --git a/hugo/content/en/llm_observability/investigate/metrics.md b/hugo/content/en/llm_observability/investigate/metrics.md index 891b639ef0d..b402a3903ed 100644 --- a/hugo/content/en/llm_observability/investigate/metrics.md +++ b/hugo/content/en/llm_observability/investigate/metrics.md @@ -21,7 +21,7 @@ further_reading: After you instrument your application with Agent Observability, you can access Agent Observability metrics for use in dashboards and monitors. These metrics capture span counts, error counts, token usage, and latency measures for your LLM applications. These metrics are calculated based on 100% of the application's traffic.
-The ml_obs.* entries on this page are Datadog Metrics: numerical values that describe an aspect of your LLM application over time, derived from your LLM spans (counts, distributions of cost, tokens, latency, errors). They are 100%-sampled, follow standard Datadog metric retention (15 months at full granularity), and are queryable from dashboards, monitors, and notebooks like any other Datadog metric. +The ml_obs.* entries on this page are Datadog Metrics: numerical values that describe an aspect of your LLM application over time, derived from your LLM spans (counts, distributions of cost, tokens, latency, errors). They are 100%-sampled, follow standard Datadog metric retention (15 months at full granularity), and are queryable from dashboards, monitors, and notebooks like any other Datadog metric.

They are distinct from two other things in Agent Observability: