Datadog AI Agents: APIs, Use Cases and Governed Automations
Understand Datadog's native AI, eight verified API operations, and how Major adds stateful incident apps, cross-system judgment, and governed approval workflows.

Datadog AI agents investigate telemetry and help engineers understand incidents. This guide separates Datadog's own AI features from agents on Major, then covers the APIs, connection method, approval controls, and cross-system automations a platform team can build.
Major keeps repeatable collection and execution in deterministic apps, while agents judge relevance and draft recommendations. Datadog remains the observability system of record.
What Datadog does for engineering teams
Datadog brings infrastructure monitoring, application performance data, logs, and operational events into one observability platform. Its API overview exposes those resources so software can retrieve evidence instead of making an engineer assemble every incident packet manually.
SREs use monitors to detect threshold breaches and investigate affected services. Application engineers inspect performance and error evidence around a deployment. Platform teams maintain monitoring coverage and service ownership, while incident commanders need a timeline they can trust across alerts and incident records.
The monitor reference describes monitors as watching a metric or check and notifying a team when a defined threshold is exceeded. That notification is a useful starting point, but it is not a complete explanation of business impact or an authorization to change production.
Major fits where the work extends beyond telemetry: joining a Datadog monitor to a PagerDuty incident, persisting the evidence gathered, and routing a proposed update to an accountable reviewer. The app holds the investigation record, the agent interprets it, and the workflow controls what happens next.
Datadog AI agents and built-in investigation
Datadog already has substantial native AI. Its Bits AI documentation names Bits Investigation for alert investigations, Bits Code for code fixes, Bits Security Analyst for threat-signal triage, and Bits Chat for exploring observability data. These are Datadog capabilities, not applications built on Major.
Watchdog continuously analyzes observability data for anomalous behavior. Its documented features include proactive alerts, investigation insights, root cause analysis, and user-impact analysis. These features are useful when the question is which telemetry signals deserve attention or what might explain a degradation.
An agent on Major adds a different layer: cross-system judgment tied to durable software. It can compare the telemetry with an incident's ownership, a release record, and the team's communication policy, then draft the next useful action. That judgment belongs to the agent, not the app.
Use Datadog's investigation capabilities where they fit. Use Major to build the persistent review queue, evidence history, and approved execution around a broader process. Do not assume that availability of a Bits feature implies a public API for it or permission to invoke it through a connector.
How Datadog connects to Major
Datadog connects through Major's Pipedream connector, app identifier p_datadog. There is no native Datadog connector. The connector catalog and this hub's connection record are the source of truth for that distinction.
IT sets up the connection once and keeps credentials centrally in the connector vault, never in app code or agent prompts. Choose the Datadog site that owns the data. The US1 API base is https://api.datadoghq.com; the official API documentation provides the other site addresses.
Configure separate access for collecting evidence and making approved changes. Before designing around an operation, verify that it is exposed through the installed connector. Public REST availability does not mean every operation is already packaged as a connector action.
Major orchestrates Datadog; it never replaces it. Datadog retains monitors and telemetry, while the Major app stores linked incident IDs, collection status, drafts, approvals, and execution outcomes. SSO, RBAC, and audit logs cover the apps, agents, and workflows using the connection. That gives IT a control point without giving each workflow author their own unmanaged secret.
Datadog API guide for apps, agents, and workflows
Authentication, pagination, and operational limits
Datadog's key documentation distinguishes organization API keys from application keys associated with a user or service account. API examples use DD-API-KEY and DD-APPLICATION-KEY headers. Prefer a service-account application key scoped to the required permissions rather than inheriting a person's full access.
The linked endpoint references specify the permissions: monitors_read for monitor reads and validation, timeseries_query for metric queries, logs_read_data for log search, and events_read for events. An approved monitor edit needs monitors_write or the documented draft-write alternative. Check the actual operation and key permissions rather than treating authentication as authorization.
Pagination differs by resource. Monitor listing uses page and page_size. Log search uses cursor and limit inside the page object; the pagination guide shows carrying the returned cursor into the next request. Event listing uses page[cursor] and page[limit]. Save progress and label incomplete collections explicitly.
Rate limits are endpoint-dependent, not one universal requests-per-minute allowance. Watch X-RateLimit-Limit, X-RateLimit-Period, X-RateLimit-Remaining, X-RateLimit-Reset, and X-RateLimit-Name. A 429 should pause collection and trigger bounded backoff using the reset information, not a fresh burst of agent calls.
The Webhooks integration sends notifications when named in monitor messages. It supports custom headers, but its public page does not document a standard cryptographic signature scheme. Configure receiver authentication explicitly and verify it before accepting a payload. The documented request timeout is 15 seconds; missed connections are retried five times, and internal notification errors or endpoint 5XX responses can trigger retries.
No general-purpose API sandbox is documented in these references. Test with nonproduction telemetry and dedicated test monitors under restricted credentials. A monitor-validation request checks configuration, not the safety or effectiveness of a proposed production change.
GET /api/v1/monitor
List monitors from the organization to establish coverage and ownership context.
- App: Store a monitor inventory with source IDs and refresh timestamps.
- Agent: Identify which monitors are relevant to a service investigation.
- Workflow: Refresh the inventory on a schedule and persist pagination progress.
GET /api/v1/monitor/{monitor_id}
Retrieve one monitor's configuration and details before explaining an alert.
- App: Show the source configuration beside the investigation draft.
- Agent: Interpret the query and notification text in incident context.
- Workflow: Fetch the current record when an alert arrives and before an approved edit.
GET /api/v1/monitor/search
Search and filter monitors to narrow the candidate evidence set.
- App: Present a filtered collection of monitors for a service owner.
- Agent: Assess whether related monitor matches support one incident narrative.
- Workflow: Run bounded searches using approved service and environment filters.
GET /api/v1/query
Query metric time series over a chosen interval.
- App: Save query parameters, timestamps, and returned evidence for comparison.
- Agent: Compare the alert window with a baseline and describe uncertainty.
- Workflow: Fetch fixed before-and-after windows without inventing query syntax each run.
POST /api/v2/logs/events/search
Search logs using a bounded filter and time window. Despite using POST, this operation retrieves evidence.
- App: Retain permitted excerpts and links, with retention controls.
- Agent: Group recurring error patterns and distinguish symptoms from hypotheses.
- Workflow: Apply query limits, follow cursors, and stop at the evidence budget.
GET /api/v2/events
List events to enrich an investigation's timeline.
- App: Store event IDs and timestamps alongside the incident record.
- Agent: Assess whether a nearby event is relevant without declaring causation from timing alone.
- Workflow: Collect events within the incident window and deduplicate across pages.
POST /api/v1/monitor/validate
Validate a proposed monitor definition without creating that monitor.
- App: Display validation results beside a proposed configuration diff.
- Agent: Explain validation failures and revise its recommendation.
- Workflow: Require validation before sending a proposed change for approval.
PUT /api/v1/monitor/{monitor_id}
Edit a specified monitor, a consequential operation that needs a separate approval boundary.
- App: Store the exact approved diff and resulting source record.
- Agent: Propose a justified change, never authorize its own production edit.
- Workflow: Recheck current configuration, enforce approval, execute once, and read back the result.
Datadog automation use cases on Major
Datadog and PagerDuty incident briefs
This is grounded in the published incident-triage use case, where the operational job is assembling context before an on-call engineer investigates.
- App: Maintain the monitor-to-incident mapping and the evidence-linked brief history.
- Agent: Correlate related alerts and draft likely affected services with uncertainties.
- Workflow: Start from an incident event, gather monitor details and metric windows, then route the brief under the team's posting policy.
Use the linked monitor-details, monitor-search, metric-query, and event-listing operations above. PagerDuty remains responsible for the incident lifecycle. See the PagerDuty automation hub for that system's role.
Datadog, GitHub, and Slack release review
A release review asks whether a deployment deserves investigation, not whether two nearby timestamps prove a regression.
- App: Hold the release identifier, evidence windows, and reviewer disposition.
- Agent: Compare GitHub release context with metric and log evidence, then draft a qualified explanation.
- Workflow: Collect approved service queries after a release and route concerning findings to a Slack review channel.
Use metric queries, log search, and event listing. Keep rollback authorization separate from the draft. The GitHub automation hub covers the repository side of the process.
Datadog and Jira monitoring follow-up
Recurring noisy alerts often need an owned engineering task rather than another automatically created ticket.
- App: Track recurring monitor findings, existing task links, and dispositions.
- Agent: Judge whether the pattern merits investigation, a monitor adjustment, or no action.
- Workflow: Run a scheduled monitor review, deduplicate existing Jira work, and request approval for a proposed ticket or configuration change.
Use monitor listing and search for candidates, then monitor details and validation for a proposed adjustment. Only the separately approved edit operation changes Datadog. This avoids letting an agent silence the evidence that prompted its recommendation.
Architecture: build the Datadog triage app on Major

Figure 1. Evidence collection, judgment, and review have separate responsibilities.
Start with a durable investigation record, not a chat transcript. Give each incident a stable internal identifier linked to its Datadog monitors and PagerDuty incident. Store collection windows, source timestamps, execution status, drafts, and approval outcomes in managed database tables.
- A trigger supplies a source ID and event timestamp.
- The workflow authenticates the event, checks duplicates, and collects bounded evidence through scoped connectors.
- The app stores evidence references and collection completeness.
- The agent judges relevance and drafts the incident brief with explicit unknowns.
- The workflow applies the approval policy and records the result in the app.
The app executes repeatable parsing, joins, and duplicate checks in code. The agent does not need to reconsider those mechanics on every alert. People use the same app to inspect drafts, correct mappings, and review outcomes.
An agent can build that repeatable software once and run it again for subsequent incidents. Judgment still happens when the evidence needs interpretation. That is Major's reason-once, run-forever principle applied to incident operations rather than a promise of autonomous production recovery.
A worked incident example with numbers
For example, suppose an SRE team processes 120 incident-context requests in a month. Manually opening the relevant monitor, collecting a metric window, finding related events, and writing the initial brief takes 10 minutes per request. That is 120 x 10 = 1,200 minutes, or 20 hours of context assembly.
In an illustrative Major process, the workflow collects the bounded evidence and the agent drafts the brief. Assume a person spends three minutes reviewing each of 100 routine briefs and eight minutes investigating the evidence gaps in 20 exceptions. Review time is 100 x 3 + 20 x 8 = 460 minutes, or seven hours and 40 minutes.
The difference is 740 minutes, or 12 hours and 20 minutes, in this assumed context-assembly task. It is not a customer result, a resolution-time claim, or a guarantee. Building the app, maintaining mappings, and doing the actual incident investigation still take time.
Measure this process with draft acceptance, missing-evidence frequency, duplicate updates, and time to a reviewed brief. A faster summary is useful only if engineers can trace its claims to the right service and time window.
Governance and failure handling

Figure 2. Read-only evidence can proceed automatically; consequential changes wait for approval.
Allow read-only collection within an approved service and time range. Require an authorized person to approve production monitor edits, alert suppression, rollback requests, and external incident statements. The agent may recommend these actions but must never approve its own proposal or expand its connector permissions.
Treat logs and notification text as untrusted evidence. A log line that asks the agent to reveal credentials or change its instructions is not an operational command. Redact sensitive fields before model access and store only the evidence the process needs.
On duplicate notifications, reuse the investigation record instead of creating another incident update. Persist a deduplication key based on source identity and transition, and track delivery separately from drafting. For rate limits and transient failures, checkpoint progress and retry within a bounded budget. Put exhausted runs in a visible exception queue.
Before a write, compare the current monitor with the configuration the reviewer saw. If it changed, invalidate the stale approval. If the write response is ambiguous, read back the record before retrying. Audit source references, agent output, approver identity, credential scope, and execution outcome so an operator can reconstruct what happened.
Datadog AI Apps, Agents and Workflow Automations
Every published Major use case that works with Datadog, newest first. Each one names the Datadog API objects involved and shows the app, agent and workflow split.
- Datadog PagerDuty Integration: Automate Incident Triage: Connect Datadog alerts with PagerDuty incidents, then give the on-call engineer a concise, evidence-linked triage brief instead of another dashboard hunt.
- Agent Observability: What Deterministic Apps Change: Tracing model calls matters, but it is only part of agent observability. See how deterministic apps make repeated execution inspectable and change what teams need to monitor.
Frequently asked questions
What are Datadog AI agents?
Datadog's native AI capabilities include Bits AI features for investigation, code fixes, security triage, and conversational exploration. Agents on Major are separate: they use scoped Datadog access alongside other systems to interpret evidence and recommend actions, while apps retain process state.
Is the Datadog Agent the same as an AI agent?
No. The Datadog Agent is software that collects monitoring data from infrastructure and applications. An AI agent performs judgment tasks such as interpreting an incident's evidence. Do not confuse installing the collection Agent with authorizing an AI investigation workflow.
Does Major have a native Datadog connector?
No. Datadog connects through Major's Pipedream connector. IT configures and scopes the connection centrally; verify that the installed connector exposes the operations your process needs.
Can an agent change Datadog monitors automatically?
The API supports monitor edits, but production changes should require an authorized approval. Keep read access separate from write access, validate the proposed configuration, and recheck the current monitor before applying the exact approved change.
Does Datadog provide an API sandbox?
No general-purpose API sandbox is documented in the references used here. Use nonproduction telemetry, dedicated test monitors, and restricted credentials. Configuration validation does not simulate the operational impact of changing an alert.
How should Datadog webhook retries be handled?
Authenticate notifications, persist them quickly, and deduplicate before downstream actions. Datadog documents a 15-second request timeout, five retries for missed connections, and retries for internal notification errors or endpoint 5XX responses. Keep your own bounded recovery and exception queue.
Sources
The following official references support the API paths, authorization model, native AI descriptions, and operational constraints in this guide. Endpoint-specific pages are linked beside each operation above so an implementer can check request fields before configuring a connector action. The pagination and webhook guides matter separately because collection completeness and delivery behavior are not interchangeable with successful authentication. Read the version-specific operation before changing permissions, and confirm your site's host before testing. A successful request against the wrong organization can still produce an incomplete or misleading incident packet.
- Datadog API overview
- Authentication reference
- API and application keys
- Monitor API overview
- Get all monitors
- Get monitor details
- Search monitors
- Query metric time series
- Search logs with POST
- List events
- Validate a monitor
- Edit a monitor
- Rate limits and response headers
- Log pagination guide
- Webhooks integration
- Bits AI features
- Watchdog capabilities
- Datadog Agent documentation
Start building on Major
Build a Datadog incident-context app with a persistent evidence history, an agent that drafts the brief, and an approval workflow that keeps production changes accountable.
About the author
Ritvij Gautam is Head of Growth at Major.
Last reviewed November 20, 2026 by Ritvij Gautam.
Frequently asked questions
- What are Datadog AI agents?
- Datadog's native AI capabilities include Bits AI features for investigation, code fixes, security triage, and conversational exploration. Agents on Major are separate: they use scoped Datadog access alongside other systems to interpret evidence and recommend actions, while apps retain process state.
- Is the Datadog Agent the same as an AI agent?
- No. The Datadog Agent is software that collects monitoring data from infrastructure and applications. An AI agent performs judgment tasks such as interpreting an incident's evidence. Do not confuse installing the collection Agent with authorizing an AI investigation workflow.
- Does Major have a native Datadog connector?
- No. Datadog connects through Major's Pipedream connector. IT configures and scopes the connection centrally; verify that the installed connector exposes the operations your process needs.
- Can an agent change Datadog monitors automatically?
- The API supports monitor edits, but production changes should require an authorized approval. Keep read access separate from write access, validate the proposed configuration, and recheck the current monitor before applying the exact approved change.
- Does Datadog provide an API sandbox?
- No general-purpose API sandbox is documented in the references used here. Use nonproduction telemetry, dedicated test monitors, and restricted credentials. Configuration validation does not simulate the operational impact of changing an alert.
- How should Datadog webhook retries be handled?
- Authenticate notifications, persist them quickly, and deduplicate before downstream actions. Datadog documents a 15-second request timeout, five retries for missed connections, and retries for internal notification errors or endpoint 5XX responses. Keep your own bounded recovery and exception queue.