Start with the systems an attacker would use
A stolen sign-in, a leaked key, and a setting changed in a hurry are the usual ways into a startup. Each one leaves a trace in a system you already run, and those systems already keep the logs. Security monitoring means collecting those logs, watching them at all hours, and acting on the few events that matter.
- Google Workspace or Microsoft 365: who signed in, from where, and what admins changed.
- AWS, Google Cloud, or Azure: who changed permissions, opened the network, or made storage public.
- GitHub: leaked secrets, repositories made public, and branch protection removed.
- Vercel, Supabase, Railway, and Cloudflare: who read or changed secrets, who joined, and what was opened to the internet.
- Laptops and servers: new admin accounts, suspicious commands, and SSH sign-ins.
What to log first
Turn on the audit logs your providers already keep, then keep them somewhere longer than the provider does. Some keep very little: Supabase keeps logs for 1 day on its Free plan, and Railway keeps audit logs for 48 hours on its Free, Trial, and Hobby plans, according to Jutsu's Supabase guide and Railway guide.
For a typical SaaS startup, this order covers the most risk for the least work:
- The place everyone signs in: Google Workspace or Microsoft 365.
- The cloud account that runs the product: AWS CloudTrail, Google Cloud audit logs, or the Azure Activity Log.
- The code: GitHub's audit events and its secret scanning, code scanning, and Dependabot alerts.
- Where you deploy and store data: Vercel, Supabase, Railway, and Cloudflare.
- The laptops and servers your team uses.
What to alert on, and what to leave as logs
Alert on the events that need a person today. Keep everything else as logs, for search and for your audit. The table below gives one example per system; these are the kinds of events worth waking someone for.
Routine events, such as a normal sign-in or a redeploy, stay out of the alert queue. In Jutsu, new connections start with a Medium-and-above severity floor: lower events are kept for search but don't raise alerts. When a source goes quiet, the rules that read it say so, and Pipeline Health shows when each source last sent events.
- A sign-in after many failures, or from two places too far apart to travel between.
- 2-Step Verification turned off, or a new super admin.
- A live credential found by GitHub secret scanning, or a private repository made public.
- A security group opened to the world, or an S3 bucket made public.
- A Supabase database opened to the whole internet, or MFA turned off for a member.
24/7 security monitoring without a security team
Logs nobody reads don't protect you. The hard part is the review: deciding at 3 a.m. whether a sign-in from a new country is your CTO on a trip or someone using their password. Jutsu's AI agents take that first look.
- Each alert gets a category, severity, risk score, and verdict.
- IP addresses, domains, and file hashes are looked up in threat intelligence feeds.
- Related alerts become one incident, so you read one story instead of ten alerts.
- When the AI isn't sure, it asks a person instead of acting.
- When an alert is real, Jutsu proposes the fix, such as blocking an IP or disabling a user. On the Startup plan a person approves it first, and every action can be undone.
Software, not a team you rent
Jutsu is software with AI agents. Nobody at Jutsu watches your alerts, so someone on your team still owns the decisions and approves responses. If you want people on call for you at night, a managed detection and response (MDR) service is the other route. SIEM for startups compares the options, with their published prices.
What it costs
The Free plan costs $0 with no credit card. It covers 5 assets and 50 AI investigations a month, with limited AI triage. It connects 1 data source, so start with the one everyone signs in through. The Startup plan connects 8 and the Growth plan 20.
Paid plans are published and billed monthly or annually: Free ($0/month), Startup ($149/month), Growth ($499/month), Enterprise (custom quote). See the pricing page for what each includes.
Where this stops
Jutsu sees only what you connect. It doesn't scan your code or dependencies itself (GitHub's own alerts flow in), and it doesn't cover prompt injection or other risks in the AI models your product uses: for those, start with the OWASP Top 10 for LLM applications. Monitoring also doesn't replace the basics: multi-factor authentication everywhere, few admins, patching, and backups.
