First-time setup checklist

The minimum configuration to run your first useful analysis, plus optional integrations that significantly improve diagnosis quality

After signup, this is the order to set things up. The first two items are required for analyses to work. The rest are optional, but each one materially improves diagnosis quality.

Required steps

These are the bare minimum. Without them, analyses cannot run.

Connect at least one log source

Without a log source, an analysis has nothing to read. Pick the aggregator your team already uses and connect it. Each integration is two to five minutes of credential setup.

GCP Cloud Logging

Service account credentials.

After saving any log source, click Test connection to confirm credentials and preview a few sample log lines.

Verify your project name and slug

Every account starts with a default project. Rename it if you'd like under Project → General. The slug becomes part of the URL and chat commands, so pick something short and lowercase: prod, api, or backend work well.

Strongly recommended

You can run analyses without these, but the diagnoses will be noticeably shallower.

Connect a code repository

Without a repo, an analysis returns findings based on logs alone. With one connected, the AI can actually read your source files and pull request diffs — turning generic suggestions into specific ones like "the regression appears to be in PR #482, which changed how expired auth tokens are handled in services/billing/charge.ts".

GitHub

Install the SigSentry GitHub App, or use a personal access token.

GitLab

OAuth (recommended) or a personal access token.

Bitbucket

OAuth (recommended) or an app password.

You'll also configure service mappings — which service name in your logs maps to which repo path.

Add custom AI context for the project

Under Project → General, you'll find an AI Analysis Context textarea. A short paragraph here is the single biggest lever on diagnosis accuracy — it tells the AI about your stack, critical paths, and known quirks before it ever reads a log line.

A working example:

Next.js 14 frontend on Vercel, Node 20 API on AWS ECS Fargate,
PostgreSQL via Neon, Redis via Upstash. Auth via Clerk.

Critical paths: /api/checkout, /api/webhooks/stripe.

Known quirk: Stripe occasionally returns 429 with rate-limit headers —
treat as warning, not error.

Deploys via GitHub Actions on merge to main, ~10 deploys/day.

The textarea is capped at 2,000 characters. See AI analysis context for best practices and patterns that work well.

Optional but high-value

Adding any of these changes how SigSentry shows up in your team's day-to-day workflow.

A notification channel

So that high-severity diagnoses are delivered automatically, and so your team learns about analyses they didn't personally trigger.

Each channel has a severity threshold, so you can route only what matters to noisy shared channels. See Notification Channels.

A chat trigger

So your team can run analyses without leaving Slack, Teams, Discord, or Google Chat. Once installed, anyone in an authorized channel can run /sigsentry analyze <description> last 30 minutes and the diagnosis posts back as a threaded reply.

See Chat Triggers.

A watchdog rule

For proactive monitoring — fire alerts when error counts or rates cross a threshold, optionally auto-running an analysis. See Watchdog.

A support desk integration

If your support team uses Zendesk, Freshdesk, or Intercom, every inbound ticket gets a severity classification automatically — with the option to auto-trigger a full analysis on high-severity tickets. See Support Desk.

Invite the rest of your team

Under Team, invite your engineering, on-call, and support folks. Match the role to the responsibility:

Role	Best for
Admin	Engineering managers, tech leads
Member	Engineers (most common choice)
Viewer	Stakeholders who only need to read

Smoke-testing what you've set up

Once you've connected at least a log source, run a test analysis to confirm the pipeline. Navigate to your project, click Analyses → New analysis, give it any description, set the time range to Last 1 hour, and submit.

If you get a valid result back, the pipeline is working.

"No logs found" usually means the time window contains no log activity, or the log source is configured for the wrong environment. Try widening the window. If it persists, see Troubleshooting.

When you're ready to walk through it end to end, the Quick start takes about five minutes.