What is PagerDuty?
PagerDuty pages the right engineer when something breaks, runs on-call schedules and escalation, and coordinates the response in Slack or Microsoft Teams. Its paid suite, renamed the PD Reliability Platform on 3 November 2026, adds noise reduction, automation, status pages and AI agents on top.
Paige, formerly the SRE Agent and generally available since the same date, is the AI feature. When an incident fires it pulls logs, metrics and traces from tools such as Datadog, Grafana, New Relic, Splunk and AWS CloudWatch, searches runbooks in Confluence, Notion or GitHub, checks recent code changes and similar past incidents, and returns likely causes with its evidence and a confidence level. It can draft a playbook, suggest a remediation workflow and, if you allow it, run the fix. Every resolved incident goes into a memory scoped to that PagerDuty service.
Paige is included in all four PD Reliability Platform tiers and sold as an add-on on the per-user Professional plan. Each request to it uses 4 AI Actions from the plan's allowance. Putting Paige on call as a virtual responder that starts investigating before a person, and using it in Microsoft Teams, are still in Early Access.
How much does PagerDuty cost?
| Plan | Price | What you get |
|---|---|---|
| Free | $0 | Up to 5 users; no AI features |
| Professional | $21/user/month annual ($25 monthly) | Paige, PagerDuty Advance and AIOps are paid add-ons |
| Starter | $2,800/year | 5 seats plus 5 bonus, up to 30; 1,000 AI Actions; Paige included |
| Essential | $12,000/year | 25 seats; 5,000 AI Actions |
| Plus | $48,000/year | 50 seats; 16,000 AI Actions |
| Ultimate | Custom | 500 seats; 160,000 AI Actions |
Free: $0 for up to 5 users, no AI. Professional: $21/user/month billed annually ($25 monthly); Paige, PagerDuty Advance and AIOps are add-ons at prices not published. PD Reliability Platform, paid upfront yearly, with Paige included: Starter $2,800/year (5 seats plus 5 bonus for a limited time, up to 30; 1,000 AI Actions), Essential $12,000/year (25 seats; 5,000 AI Actions), Plus $48,000/year (50 seats; 16,000 AI Actions), Ultimate custom (500 seats; 160,000 AI Actions). Each Paige request uses 4 AI Actions; extra actions are bought with credits at a price not published.
Is PagerDuty worth it?
The worst part of a 3am page is the start. You open the dashboard, then the logs, then the deploy history, then the runbook nobody updated, trying to work out what changed before you can even start fixing it.
Paige does that legwork. It reads the alert, queries logs and metrics in tools like Datadog, Grafana or CloudWatch, checks recent pull requests and similar past incidents, and posts a short list of likely causes with the evidence behind each. PagerDuty already sits at the centre of the incident for most teams that use it, which is the real advantage over a stand-alone AI agent: it already holds the alert, the service it hit and what happened last time that service broke. PagerDuty also claims, from its own benchmark, that Paige beat leading AI coding agents at diagnosing simulated incidents; that is not an independent test.
It connects to 22 tools, from Datadog and Splunk to Sentry, GitHub and Notion. An admin enters an API key or signs in for each one, and every tool Paige could call is listed as read-only or write, each set to Always Allow or Blocked. Fixes wait for your approval unless you decide otherwise, and permissions can be scoped per team.
The catch: the price of Paige is hard to pin down. On the Professional plan, at $21 per user a month, it is an add-on with no published price, and using it in the Operations Console also needs the separate AIOps add-on. On the PD Reliability Platform it is included, but every question costs 4 AI Actions, so Starter's 1,000 actions at $2,800 a year cover about 250 Paige requests, and PagerDuty does not publish what more actions cost. Credits expire each year. Access is the other cost. Paige needs keys for your log and monitoring tools, and PagerDuty does not publish the minimum scopes for each one, so keeping those keys read-only is your job. It reads only the first 2,000 characters of an alert's details and each note, and it saves what it learned only after the incident is resolved.
On data, AI features are opt-in, PagerDuty says no customer data is currently used for training, and its Slack listing names Claude on AWS Bedrock, Azure OpenAI and IBM watsonx.ai as the models, configured to retain no data.
Our take: if PagerDuty already wakes your team, Paige is worth a trial on one noisy service, with read-only keys and approvals on. A team choosing an incident tool from scratch should get the AI Actions price in writing before signing.
Where it earns its place
An on-call SRE at a 150-person fintech running Datadog and GitHub
Know what broke before I open five tabs at 3am
Connect Datadog logs and GitHub code with read-only keys, then ask Paige in the Slack incident channel; it returns likely causes with evidence and links to similar past incidents. Each question uses 4 AI Actions from the plan's allowance.
A head of platform at a 60-engineer SaaS company on the Essential tier
Let the agent handle low-priority alerts without letting it touch production unsupervised
Add Paige to an escalation policy for low-urgency services, block every write tool except GitHub's revert-PR action, and require approval for remediation. Essential at $12,000 a year includes 5,000 AI Actions, about 1,250 Paige requests; the virtual-responder role is Early Access.
A security officer approving AI access to production logs
Decide what Paige may read and where that data goes
Issue dedicated read-only API keys per connector, scope Paige's agent profile to one team, and review the AI Disclosure on PagerDuty's assurance portal. Paige's memory can be viewed, corrected or redacted through the Memory API.
Before you commit
- Setup
- An Admin or Account Owner opens AI Settings, enables Paige, connects each tool on the Connectors tab with an API key or OAuth, and sets each tool to Always Allow or Blocked; Slack needs re-authorising for extra scopes.
- Integrations
- 22 connectors: logs, metrics or traces from Datadog, Grafana, New Relic, Splunk, Dynatrace, Coralogix, Honeycomb, Sumo Logic, Elasticsearch, Logz.io, Sentry, Observe, Arize, Langfuse, AWS CloudWatch, Azure Monitor and GCP; runbooks from Confluence, Atlassian, Notion, ServiceNow and GitHub; GitHub code search and revert PRs. Works in Slack and the web app; Teams is Early Access; more than 750 integrations feed the wider platform.
- Data and privacy
- AI is opt-in and can be switched off. PagerDuty says no customer data is currently used for training; its Slack listing names Claude on AWS Bedrock, Azure OpenAI and IBM watsonx.ai, configured to retain no data. Paige's memory is scoped per PagerDuty service and can be redacted through the Memory API. Atlassian and Notion access is read-only. Which AI tools train on your data?
- Cost at scale
- Starter's 1,000 AI Actions cover about 250 Paige requests, Essential's 5,000 about 1,250 and Plus's 16,000 about 4,000. Extra actions are bought with credits at a price PagerDuty does not publish; accounts signing Essential or above by 15 January 2027 get extra actions free for the first year.
- Limits
- Reads only the first 2,000 characters of custom details and each note; memory is saved only after an incident is resolved; GitHub runbooks must be on the main branch; Atlassian searches spend Rovo credits; Splunk APM is not supported.
What’s good
- Reads logs, metrics, traces, code and runbooks across 22 connectors
- Every tool can be set to Always Allow or Blocked, with write tools separated
- Fixes wait for human approval by default; permissions scoped per team
- Included in every PD Reliability Platform tier from $2,800 a year
- Opt-in AI; PagerDuty says no customer data is used for training
What’s not
- Each Paige request uses 4 AI Actions; extra actions have no published price
- Add-on on Professional at a price not published
- Virtual responder and Microsoft Teams support are still Early Access
- Minimum read-only scopes for each log connector are not documented
- Benchmark claims come from PagerDuty's own report
Best for
SRE and platform teams already paging through PagerDuty who want first-pass incident investigation without adding another vendor.
PagerDuty: common questions
- Is PagerDuty worth it?
- Paige turns PagerDuty from the thing that wakes you up into the thing that has already read the logs, but you pay for it in AI Actions you cannot price and log access you have to scope yourself. We rate it 4.0 out of 5.
- How much does PagerDuty cost?
- Free: $0 for up to 5 users, no AI. Professional: $21/user/month billed annually ($25 monthly); Paige, PagerDuty Advance and AIOps are add-ons at prices not published. PD Reliability Platform, paid upfront yearly, with Paige included: Starter $2,800/year (5 seats plus 5 bonus for a limited time, up to 30; 1,000 AI Actions), Essential $12,000/year (25 seats; 5,000 AI Actions), Plus $48,000/year (50 seats; 16,000 AI Actions), Ultimate custom (500 seats; 160,000 AI Actions). Each Paige request uses 4 AI Actions; extra actions are bought with credits at a price not published.
- What are the downsides of PagerDuty?
- Each Paige request uses 4 AI Actions; extra actions have no published price; add-on on Professional at a price not published; virtual responder and Microsoft Teams support are still Early Access; minimum read-only scopes for each log connector are not documented; benchmark claims come from PagerDuty's own report.
- Who is PagerDuty best for?
- SRE and platform teams already paging through PagerDuty who want first-pass incident investigation without adding another vendor.
- How much does PagerDuty cost for a team or in production?
- Starter's 1,000 AI Actions cover about 250 Paige requests, Essential's 5,000 about 1,250 and Plus's 16,000 about 4,000. Extra actions are bought with credits at a price PagerDuty does not publish; accounts signing Essential or above by 15 January 2027 get extra actions free for the first year.
- Is PagerDuty safe for company data?
- AI is opt-in and can be switched off. PagerDuty says no customer data is currently used for training; its Slack listing names Claude on AWS Bedrock, Azure OpenAI and IBM watsonx.ai, configured to retain no data. Paige's memory is scoped per PagerDuty service and can be redacted through the Memory API. Atlassian and Notion access is read-only.
Answers checked Nov 16, 2026.