Datadog, questioned in plain words.
Ask what spiked at 3am, which monitors are firing, how much error budget is left. Cedric reads Datadog and proposes any change before running it.
$50 in free credits when you sign up. No card needed.
- Datadog has the graph; the question in the channel is what happened and whether it is still happening.
- Ask in plain words: the p95 latency since the deploy, the errors in the checkout logs, every monitor firing right now.
- Cedric checks service level objectives and error budgets, and reviews incidents and their timelines.
- Monitors can be created, updated, muted or deleted, maintenance windows scheduled, and dashboards built or updated, each on your yes.
- Deleting a monitor or a dashboard and muting alerts change what the team sees, so they are confirmed first, and retention limits how far back logs and metrics go.
What can a Datadog AI agent do?
Most questions about production are questions about Datadog. A Datadog AI agent queries metric data over a time range to check trends, spikes or system health, and searches and reads logs and events to investigate an error or a recent deployment. It lists, inspects and searches alert monitors with their current status, so "what is firing" gets a list, and looks up monitored hosts, services and tags, including which hosts are reporting or muted. Service level objectives, error budgets and incidents with their timelines are readable too.
Changes come as proposals. Cedric creates, updates, mutes or deletes monitors and schedules maintenance windows that suppress alerts, so a planned migration does not page anyone. It builds, retrieves, updates and shares dashboards and their individual widgets, which is how a dashboard for the new service can come out of a thread. It also submits custom metrics and posts events, so a deploy or a marketing launch shows up on the timeline next to the monitoring data. That is the monitoring end of an AI operations assistant.
A few things to know. Per-endpoint rate limits apply, so a wide log search is paged. Historical log and metric data is limited by your retention settings: Cedric can only read as far back as Datadog keeps. Destructive writes, deleting monitors or dashboards or muting alerts, are confirmed before they run, because a muted alert is one nobody sees.
3am, read at 9
The spike, the logs, and the maintenance window.
Nadia Petrova9:14 AM
@Cedric what spiked on api-gateway around 3am, and set a maintenance window for the database migration tomorrow, 22:00 to 23:30, so the latency monitors don't page.
CedricAPP9:15 AM
api-gateway p95 latency went from 180ms to 2.1s between 02:58 and 03:14, error rate from 0.2% to 6%. The logs show "connection pool exhausted" on db-primary, 1,340 lines starting 02:57, right after the nightly export job. Two monitors fired and recovered by 03:20; the incident "API latency 3am" is open with no timeline yet. Proposed: a maintenance window tomorrow 22:00 to 23:30 on the 4 latency monitors tagged service:api-gateway, named "DB migration", and the pool exhaustion noted on the incident timeline.
Queried
Metrics over a time range, and logs and events searched for the error or the deploy behind a spike.
Listed
Monitors with their current status, hosts and services with their tags, SLOs and error budgets, incidents and their timelines.
Changed
Monitors created, updated or muted, maintenance windows scheduled, dashboards built or updated, proposed first and saved on your yes.
Connect Datadog in a minute
No setup project, no workflow builder, no field mapping. The flow is the same for every tool Cedric connects to.
- 01
Install Cedric where your team talks
One click. It joins your Slack like any other member, and everyone shares the same Cedric.
- 02
Connect Datadog from the dashboard
Open the Integrations page and sign in with Datadog on its own screen. Cedric never sees or stores your password, and you can revoke the connection in one click at any time.
A fresh connection can look but not touch: read-only until you say otherwise.
- 03
Ask in plain words
Back in the channel, try it: "which monitors are firing right now?". You describe the outcome; Cedric picks the Datadog actions the ask needs. There is nothing to configure.
Connect for the whole workspace so the team shares it, or keep the connection personal. Two accounts are fine too; each gets a short name.
- 04
Let it become routine
Schedule the recurring version of the ask, and let Cedric fold Datadog into bigger jobs that cross your other tools. Anything that would leave your workspace still comes back as a preview to approve first.
Better in company
A spike is usually a deploy: pair Datadog with GitHub and Cedric names the merge that landed minutes before the metric moved, and files the issue in Linear with the log lines attached, on your yes.
The uptime number belongs in the customer update: Cedric can read the SLO and the error budget and draft the note in Gmail or the status page in Notion, with the incident timeline summarized in plain words.
Where the limits are
Datadog is a catalog connection, not a built-in playbook. Cedric works from what the Datadog connection can read and do; ask in plain words and it says what it can reach there, and what it cannot.
Retention sets the horizon. Historical log and metric data is limited by your Datadog retention settings, and per-endpoint rate limits mean a wide search is paged rather than pulled at once.
Muting and deleting are confirmed. Muting an alert, scheduling a maintenance window, or deleting a monitor or a dashboard changes what the team sees, so Cedric names the target and asks every time.
Cedric sees what the Datadog connection grants and nothing more. Access is scoped, logged, and revocable from the dashboard, effective immediately.
Nothing sends itself. An email, a post, a change to a record: anything with consequences comes back as a preview for a human yes. That is the design, not a beta caveat.
FAQ
Yes. Datadog is one of the 3,200+ apps in Cedric's catalog: connect it once from the dashboard, on Datadog's own sign-in screen, and ask in plain words. There is no built-in playbook for it, so Cedric works from what the connection can read and do, and tells you when something is out of reach.
It can read the evidence: the metrics over the window, the logs and events around it, which monitors fired, and the incident's timeline. It says what it found and what changed just before; deciding the cause and the fix stays with the on-call engineer.
Yes, on your yes. Describe the service and the widgets you want and Cedric proposes the dashboard, or updates one that exists and shares it. Nothing is created until you approve.
Connect it for the whole workspace and everyone shares it, with one person authorising once. A connection you would rather not share, a personal inbox for instance, can stay personal to you.
Yes, and it starts that way: a fresh connection can look but not touch until you say otherwise. Even with write access granted, anything that changes or sends waits for an approval.
Every workspace starts with $50 in credits, free, no card. Each task consumes credits in proportion to the work involved. See pricing for the plans.
Connect Datadog in a minute.
Ask which monitors are firing, then let it read the logs behind the last spike.
100K credits ($50) on sign up. No card needed.