IT & Security
Answer a page, set up a new person's access, trace an error spike and gather audit evidence. Account changes, acknowledgements and posts wait for your approval. It works in PagerDuty, Datadog, Auth0 and Cloudflare, and across the 1,390 apps that connect today.
Pick a scenario to see what the agent reads, what it prepares, and what waits for your yes.
Datadog, PagerDuty and a Slack thread
Something is pushing database CPU past 90% and PagerDuty is on me about it. Acknowledge the alert, open an #incidents thread, and find out whether recent traffic lines up with what Datadog shows.
Should the runbook for the kill switch go into the #incidents thread too? That thread would open with my read: a recently added cron job is tripling reads on 'checkout_events', which has kept DB CPU at 94% since 14:38. It names you incident commander, and it waits for your approval, as does acknowledging INC-4821.
Database CPU alert
PagerDuty · acknowledge, Slack · start_thread
Ready to paste
Paste one into any agent connected to Spinrun, put in your own names, and send.
“Open an #incidents thread for the alert PagerDuty just sent, and tell me whether traffic explains what Datadog is showing.”
“Someone new is joining engineering. Their team's Slack channels, the team's Bitwarden collection and dev-tools access in Auth0: set up all three.”
“Warn #eng-payments about the jump in Sentry errors on /api/charge, with the fix you'd make and the deploy that caused it.”
“Our auditor wants SOC2 evidence for Q1: put the WAF settings from Cloudflare and the access reviews on one Notion page.”
Safe by default
An agent works inside your accounts, so approvals, sign-ins and records are handled by Spinrun rather than left to the model.
By default, every change an agent makes in a connected app waits for your approval, on every plan. Enhanced Control adds per-app read, write and destructive rules at the gateway on Free and Pro.
You sign in once through OAuth and the gateway keeps the tokens encrypted. The agent never holds a password, so there is none for a prompt to pull out.
Models are never trained on anything you connect. The log does hold each call's arguments and a preview of what came back, email or chat text included, and deletes them when your plan's window ends.
Connect each environment's accounts as their own connections. The agent acts in the one a request names, and asks when it could be more than one.
On-call and identity access sit in Spinrun. Switch models without granting anything again.
Paging, sign-in and monitoring lead the list. Need Bitwarden, Jira or Confluence? They're among the 1,390 apps that connect today, one sign-in each.
1,380 more apps connect the same way
Use cases
Anything in the toolkit catalog. Most connect through a managed OAuth app; for the rest a workspace admin registers your own OAuth app once, from the app's Set up page.
Free to start, with 10,000 one-time credits. By default, writes to your apps wait for your approval.
We would like to switch on analytics, so we can see which pages are worth keeping. Nothing has loaded yet and nothing will until you choose. The cookies that keep you signed in and remember your language are not part of this.
Read the Cookie Policy
