Skip to main content
Find the section that matches what you’re seeing and follow the steps. If your issue isn’t covered here, please contact your Rootly CSM for help.

​
The assistant isn’t responding

The bot looks installed but messages go nowhere. Work through these in order:

​
The bot replies with: "AI agent is not enabled for this team."

​
Confirm @Rootly in Slack is turned on

A Rootly admin has either not turned the assistant on, or has toggled it off.
  1. Go to Configuration → Rootly AI in Rootly
  2. Confirm Enable @Rootly in Slack is turned on
  3. Only a Rootly admin can change these.

​
Reconnect Slack

Your Slack integration with Rootly may need to be reauthenticated:
  1. Go to Configuration → Slack
  2. Look for the banner at the top indicating the Slack connection is live.
  3. Click Disconnect and then Connect Slack
  4. Complete the authentication to reconnect Slack to Rootly. This step must be run by a Slack Workspace Admin.
Updating the Slack connection does not affect your existing channels, integrations, or workflows. It re-installs the Slack app with the latest capabilities the assistant needs.
Other things to check:
  • Rootly and Slack statuses. Check the Rootly or Slack status pages for current incidents.
  • Your permissions changed. If your Rootly role was updated, the assistant may refuse actions it previously allowed.

​
Rate-limiting on your team’s environment

The Rootly assistant has a per-minute token budget. When exceeded, the assistant stops the current request and waits until the window rolls over (within 60 seconds) before accepting new requests.
  1. Wait about a minute and try again.
  2. If you’re hitting the ceiling repeatedly, talk to your Rootly admin and CSM about higher usage limits for your team.

​
”It said it doesn’t have access to that”

The assistant refused to show you something, citing permissions. The assistant runs as your Rootly user with your exact Rootly permissions. If you can’t see something in the Rootly web app, you can’t see or access it through the assistant either. What to check:
  1. Sign in to the Rootly web app and try to view the same data directly. If it’s invisible there too, your role doesn’t grant access.
  2. Check your role under Configuration → Members. Talk to your Rootly admin if you believe your role should allow it.
  3. Private incidents are only accessible from inside their own incident channel. Cross-channel and DM requests for private incident data are refused by design.
  4. If your role changed recently, sign out of Slack and back in so the bot picks up your current identity.

​
”It said it would do something but the action didn’t happen”

You asked for a write action but you don’t see the result. Writes execute as your Rootly user. If your role doesn’t allow the action — or a per-status or per-severity restriction kicks in — the action is blocked.
  • The assistant should post a plain-English message in the thread explaining why the action failed. Look for a follow-up message before assuming silent failure.
  • If the restriction is correct, ask someone with the right role to perform the action. If your role should allow it, talk to your Rootly admin.
  • Occasionally a tool call fails for a transient reason. This could be related to an upstream service hiccup or a brief timeout. The assistant should surface an error in the thread.

​
”It asked me a clarification question and I didn’t see it”

If the assistant asked you a clarifying question (e.g. "Which David did you want to page?") and you didn’t respond within 60 minutes, the card expires. Clicking an expired card shows "⌛ This question has expired.". If you don’t see the card:
  • Check that you have message permissions in that channel. Cards posted to channels you’ve left or been removed from won’t appear in your sidebar.
  • The card may have expired. Clarification cards expire after 60 minutes. If you walked away and came back later, the card is no longer actionable.
  • To fix, try re-asking the same question. A new card will be posted with a fresh 60-minute window.

​
”It looped for a while before answering”

You saw the assistant work for a noticeable stretch before posting its final reply. For complex multi-step requests, this can take several seconds and multiple internal tool calls. That’s normal. The assistant looks up real data (the right team, the right schedule, the current on-call) before taking actions, instead of guessing. When it’s not normal: if the assistant exceeds its internal step limit, it stops and posts:
I reached the maximum number of steps for this request. Try breaking it into smaller requests.
This usually means the request was too broad or ambiguous, and the assistant kept exploring without converging. To fix, try breaking your request into smaller, more specific asks related to an incident. Instead of audit all our open incidents try tell me about this incident

​
”It got a generic error”

The bot posted a message like:
⚠️ Error: <message> Something went wrong.
What’s happening: something failed below the assistant related to a provider, internal error, or a timeout. To fix, wait a moment and try again. If it keeps failing, report it to your Rootly admin or account team. Failed requests don’t automatically retry by design, to prevent retry storms during provider outages.

​
”It told me it can’t do X but I expected it could”

The assistant refused a request you thought was in scope. A few common categories of deliberate refusal:
  • Production-runtime actions. Rollbacks, restarts, deploy triggers — the assistant doesn’t run code or hit production systems directly. It can create an action item and page the on-call to run it.
  • Third-party paging. PagerDuty, Opsgenie, JSM Ops paging is by design not supported. The assistant pages through Rootly On-Call only.
  • Code generation. The assistant won’t write code or open pull requests.
  • Platform configuration changes. Adding new severities, editing custom field schemas, changing RBAC roles, SLA settings — these live in the Rootly admin UI, not in the assistant.
  • Off-topic questions. Weather, jokes, programming help, anything outside incident management. The assistant refuses with a fixed line.
  • Predictive or forecasting questions. “When will the next outage happen?” — the assistant is reactive and grounded in present and past data.
See What @Rootly in Slack can do for the full picture of what’s in and out of scope.

​
”It gave me wrong information”

If the assistant gave you a factually incorrect answer:
  1. Click Not helpful (👎) on the response.
  2. Choose a category in the “What went wrong?” modal — Hallucinated / incorrect info, Used wrong tools, Missed obvious context, Too slow / timed out, Irrelevant response, or Other.
  3. Add an optional comment describing what you expected.
Feedback flows into our managed evaluation and observability platform. It informs prompt and tool improvements over time. It does not directly retrain the model. If the issue is urgent, for example, the assistant claimed an action succeeded when it didn’t, please submit the feedback form and report it directly to your Rootly CSM.

​
”It mixed up two people / services / teams with similar names”

When you refer to a record by name, the assistant searches and picks the best match. If you have two services named “payments” and “payments-api,” or two users with similar names, the assistant should ask for clarification, but if a single match looks confident, it may proceed with the wrong one. Best practices:
  • Use exact, unique names for teams, services, and users.
  • When the assistant offers options in a clarification card, pick from the list rather than restating the name in free text.

​
Other common gotchas

  • The default rendering for times is your team’s time zone. The assistant will convert to UTC, your personal time zone, or any other IANA time zone when you explicitly ask.
  • For the time being, Ask Rootly in Web has different capabilities than @Rootly in Slack. Enhancing Rootly chat functionality on Web and Mobile are both on the roadmap.
  • Tenant isolation is enforced by design. the assistant does not surface incidents from teams you don’t belong to.

​
Known current limits

These are deliberate limits worth knowing about as you plan your deployment. For the full enterprise-readiness picture, see Enterprise readiness and security.
  • No per-user rate limit. The token budget is per-team. There’s no separate per-user ceiling.
  • BYOK is not supported yet. You can route to your own OpenAI account in Rootly Web; However, BYOK for the @Rootly in Slack is on the roadmap and not currently available.
  • No production-runtime actions. No code execution, no shell, no direct production system calls.
  • No third-party paging. Paging routes through Rootly On-Call only.
  • No retro publishing from Slack. The assistant can draft a retrospective in the incident thread; editing and publishing the retrospective document happens in the Rootly Web app.
  • No mutations to platform configuration. Severity catalogs, custom field schemas, RBAC roles, and SLA settings are managed in the Rootly admin UI, not through the assistant.