Purpose-built AI for production.

Rootly AI learns from every incident to power every agent your engineers use, so the last outage helps prevent the next one.

Book a demo Get started for free

claude · ~/checkout-api

✻ Welcome to Claude Code! · cwd: ~/checkout-api

> I'm about to lower the checkout-api pool size. Has that caused an incident?

⏺ rootly - search_incidents (MCP)(query: "checkout-api database pool")

⎿ {"data":[{"id":"7c2e19a4-5b1f-4e0a-9d3c-2f8e6a1b4c70","type":"incidents","attributes":{"parent_incident_id":null,"sequential_id":2304,"title":"Checkout latency above 4s","slug":"checkout-latency-above-4s"

… +212 lines (ctrl+o to expand)

⏺ rootly - find_related_incidents (MCP)(incident_id: "7c2e19a4-5b1f-4e0a-9d3c-2f8e6a1b4c70")

⎿ {"target_incident":{"id":"7c2e19a4-5b1f-4e0a-9d3c-2f8e6a1b4c70","resolved_incident_id":null,"title":"Checkout latency above 4s"},"related_incidents":[

… +38 lines (ctrl+o to expand)

⏺ rootly - suggest_solutions (MCP)(incident_id: "7c2e19a4-5b1f-4e0a-9d3c-2f8e6a1b4c70")

⎿ {"target_incident":{"id":"7c2e19a4-5b1f-4e0a-9d3c-2f8e6a1b4c70","resolved_incident_id":null,"title":"Checkout latency above 4s"},"solutions":[{"incident_id":"e41b7d90-28c6-4f3a-b5e2-91d0c7a63f18","title":"Pool exhaustion after config change","similarity":0.91

… +27 lines (ctrl+o to expand)

⏺ rootly - get_oncall_handoff_summary (MCP)(timezone: "America/Los_Angeles")

⎿ {"success":true,"timestamp":"2026-10-20T08:14:03+00:00","timezone":"America/Los_Angeles","schedules":[{"schedule_id":"3f6b0c2e-9a14-4d7e-8b25-c06f1e93d4a7","schedule_name":"Commerce - Primary","team_name":"Commerce"

… +64 lines (ctrl+o to expand)

⏺ Yes. INC-2304 started when deploy #1294 cut this pool from 50 to 10, and under peak load 18% of checkouts timed out for 29 minutes. INC-4102 was the same failure, and both were fixed by restoring the pool to 50. Keep it at 40 or above, and if you ship it, Dana K. is on call for Commerce until 9:00.

Every incident makes the next one less likely.

Rootly AI understands production the way your most experienced engineers do, and it remembers what every incident teaches.

checkout-api Service
- Owned byCommerce
- Depends onpostgres-main
- Callspayments-gateway
Changed 6 min ago · deploy #1294
It knows your production
Your services, what depends on what, who owns it, and what just changed, drawn from the tools you already run.
It remembers what you learn
Each investigation leaves behind what helped, so the next responder starts where the last one finished.
Checkout latency above 4s
AI SRE is investigating Auto-run
1. Gather facts
2. Triage signal
3. Draft hypotheses
It starts before you do
Investigation rules let AI SRE pick up an alert the moment it fires, so the first responder opens a working theory, not a blank page.

See the agents at work →

Tailored to your organization, in plain language.

Provide custom AI instructions to tailor AI output to the way you and your org work.

Instructions
Guidance in plain language for AI SRE and every agent, like evidence to prioritize or systems to check first.
Investigation rules
Choose which alerts AI SRE investigates automatically, add alert-specific guidance, and pause rules when needed.
Memory you curate
Keep the lessons that help, and discard the ones that don't.

Brady Lamb, Senior Manager of DevOps, SRE, IT, and Security

Confidence scores and the train of thought you can trust.

It tests competing causes, lets challenger agents push back on its answer, and shows what it ruled out, why, and how confident it is.

Agents that challenge it
Before an answer is final, challenger agents push back on it the way a coworker would, on timing, mechanism, and coincidence.
Ruled out, with reasons
Every cause it dropped stays on the page, with the reading that dropped it.
Confidence it earned
Code, not the model, sets the confidence. When nothing clears the bar, it says so instead of guessing.

AI SRE · INC-2304

Root cause identifiedHigh confidence · 2 decisive signals

Deploy #1294 cut the checkout-api database pool from 50 to 10.

Evidence

DecisivePool wait p99 up 30x from 2:10 AM Datadog
Decisivepool: 50 → 10 in #1294, merged 2:09 AM GitHub
SupportingCheckout requests queued for a connection Datadog

Memory pointed here first, from INC-4102. A note isn't evidence, so the pool was checked live.

Investigation path 3 possible causes · 9 checks · 27 tool calls

confirmedConfig change in deploy #1294The pool dropped two minutes before the onset.
ruled outDatabase CPU saturationCPU held at 31% through the window.
ruled outPayments provider latencyIts p99 stayed at 240 ms.

Challenge findings

✓Timing The change came before the onset, not just the alert.
✓Mechanism A smaller pool explains the step in pool wait.
✓Coincidence No pool waits like this in the last 30 days.

The answer held, so confidence stays High.

Santiago Suarez Ordoñez, CEO and Co-Founder

Plug Rootly into your agent stack.

Bring Rootly into Claude Code, Codex, or any AI, and plug your own tools into Rootly.

Rootly inside your AI
Query incidents, on-call, and past fixes from Claude Code, Codex, or any AI that speaks MCP.
Your tools inside Rootly
Connect your own MCP server and choose exactly which tools Rootly AI can call.

Terminal

/plugin marketplace add Rootly-AI-Labs/rootly-claude-plugin/plugin install rootly@rootly-plugins✓ Installed rootly · 9 commands

/rootly:respond/rootly:ask/rootly:oncall/rootly:retro

~/.codex/config.toml

[mcp_servers.rootly]url = "https://mcp.rootly.com/mcp"bearer_token_env_var = "ROOTLY_API_TOKEN"

.cursor/mcp.json

{ "mcpServers": {    "rootly": { "url": "https://mcp.rootly.com/mcp" }} }

Set up the MCP server →

You and your teams deserve modern incident management.

Book a demo Get started for free