Your CI broke.
PR is already open.
InariWatch monitors GitHub, Vercel, Sentry, and your own app via @inariwatch/capture. When something breaks, AI reads your code, writes the fix, waits for CI, and opens a PR. You just approve.
AI analysis included — no key neededOpen source (MIT)From error to merged PR in 2 minutes. Fully automated.
8
integrations monitored
5 min
cloud poll interval
3×
CI retry loop
5
AI providers supported
AI Remediation
Monitoring tools tell you what broke.
InariWatch fixes it.
No other monitoring tool closes the loop from alert to merged fix. Here's exactly what happens the moment something breaks.
Alert fires
CI fails, deploy errors, Sentry regression, or your own app via @inariwatch/capture — caught in real time.
AI reads your code
Connects to your repo, fetches relevant files, diagnoses the root cause from the actual stack.
Fix generated
AI writes the code change with a plain-English explanation — not a generic suggestion, an actual diff.
CI validated (with retry)
Pushes to a branch, monitors CI. If it fails, reads the logs and tries a different fix — up to 3×.
PR opened
When CI passes, opens a PR with full context. You get a notification. One click to approve.
03:12 — alert received
🔴 CI failing on main
TypeError: Cannot read 'user' of undefined
auth/session.ts:84 · build #1247
Triggered by: PR #61 merged 4 min ago
→ Starting AI remediation...
→ Reading auth/session.ts, lib/auth.ts
→ Generating fix...
→ Pushing branch fix/session-null-check
→ Waiting for CI...
→ CI passed ✓
→ PR #62 opened
03:14 — PR ready for review
fix: add null check for session.user
Root cause: PR #61 refactored the session
object but auth/session.ts still assumed
user was always defined on the response.
Changed:
- return session.user.id
+ return session.user?.id ?? null
CI: ✓ all checks passed
Waiting for your approval →
From alert to ready-to-merge PR in under 2 minutes. While you were sleeping.
Auto-merge
You sleep. We ship.
Safely.
Skeptical about AI auto-merging code into production? Fair. Here's how we make sure every auto-merged fix is something we'd sign off on ourselves.
5 gates — all must pass to auto-merge
CI required
All tests must pass before any merge is considered
Confidence ≥ 90%
Diagnosis must be clear — low-confidence fixes become draft PRs
AI self-review ≥ 70/100
A second AI call reviews the fix like a senior engineer before it ships
≤ 50 lines changed
Large or complex changes are always sent as draft PRs for human review
Auto-merge enabled
You activate this per project — off by default, always in your control
Evaluating safety gates...
✓ CI passed (3 checks)
✓ Confidence 94% ≥ 90% threshold
✓ Self-review 88/100 — approved
✓ Lines changed: 12 ≤ 50 max
✓ Auto-merge enabled
→ All gates passed — merging PR #62...
✓ Merged. Watching for regressions (10 min)
Sentry: ok Uptime: ok
✓ No regressions detected — fix is stable.
Regression detected? We revert automatically.
If Sentry catches the same error or uptime drops after a merge, InariWatch opens a revert PR and merges it — all within the 10-minute monitoring window. You wake up to a stable main branch.
Auto-merge is off by default. You enable it per project, set your own confidence threshold, and define the max diff size. Read the full Trust Architecture →
Better Together
They provide the signals. We provide the fix.
InariWatch isn't here to replace GitHub, Vercel, or Sentry. They are best-in-class at what they do. InariWatch simply connects them into a unified brain, automatically correlating their signals and finding the root cause.
The Foundation: Your Stack
Workflow failed on main
Production deploy errored
TypeError: 23 new events
The Superpower: InariWatch
🔴 Deploy failure caused by TypeError
PR #61 modified session handling → deploy failed →
TypeError at auth/session.ts:84 · 23 users affected
PR #62 ready to merge → CI ✓
Integrations
Monitors your entire stack
GitHub CI, Vercel deploys, Sentry errors, Datadog monitors, uptime, database health, dependency vulnerabilities, and your own app via the capture SDK — all in one place, already correlated.
GitHub
live- ·Failed CI checks
- ·Stale & unreviewed PRs
- ·Pre-deploy risk on PRs
Vercel
live- ·Failed production deploys
- ·Failed preview deploys
- ·Instant rollback
Sentry
live- ·New issues
- ·Regressions (resolved → reopen)
Uptime
live- ·Endpoint downtime
- ·Slow response time (configurable)
PostgreSQL
live- ·Connection failures
- ·High connections
- ·Long-running queries
npm / Cargo
live- ·Critical CVEs
- ·High-severity vulnerabilities
Datadog
live- ·Log anomalies
- ·Infrastructure spikes
- ·APM Traces
@inariwatch/capture
live- ·Exceptions from your app
- ·Custom log events
- ·Deploy markers
Platform
A complete automated monitoring platform
Alert analysis and correlation are included free. Bring your own AI key (Claude, OpenAI, Grok, DeepSeek, or Gemini) to unlock auto-fix and remediation.
AI Code Remediation
Reads your repo, writes the fix, pushes a branch, waits for CI. If CI fails, reads the logs and retries with a different approach — up to 3 times. Opens the PR only when it passes.
Pre-deploy Risk Assessment
AI reads every PR diff against your actual incident history and posts a risk score (Low / Medium / High) directly on GitHub before you merge. No config needed.
Anomaly Detection
Detects alert frequency spikes (3–5× vs baseline), repeating failure loops, integration degradation, and silent projects — proactively, before they escalate.
Alert Correlation
When Vercel fails and Sentry spikes at the same time, AI groups them into one correlated alert with a single root cause. 5 alerts become 1 actionable message.
Ask Inari
Chat with your live monitoring data. "What failed this week?", "Which integration has the most errors?", "Summarize last night's incidents." Inari has the context.
On-Call Rotations
Set up timezone-aware schedules with multi-level escalations. When a critical issue happens, InariWatch pages the specific developer on-call (and escalates if they don't respond).
Interactive Chat ACK
Acknowledge or resolve alerts directly from Telegram or Slack using interactive inline buttons. No need to open your laptop at 3 AM just to silence an alarm.
Incident Storm Control
When your database crashes, you get 1 correlated 'Incident Storm' alert instead of 50 individual service failure emails. Silence the noise when the whole stack is burning.
Uptime Monitoring
Constant 1-minute global pings to your endpoints. If your site goes down, InariWatch catches it instantly, creates an alert, and updates your Status Page.
Slack Bot
Errors arrive in Slack with AI diagnosis. Click Fix It to trigger remediation — progress, PR link, and Approve & Merge all in-thread. Ask @InariWatch questions, manage on-call, monitor deploys.
VS Code Extension
Errors appear as inline squiggly lines in your editor. Hover for AI diagnosis. Sidebar shows all alerts grouped by file. Status bar shows unread count. Works in local mode too.
Substrate I/O Recording
Capture every HTTP call, DB query, and file operation. When an error occurs, the last 60 seconds of I/O are attached to the alert. The AI sees exactly what your code did.
Community Fix Network
Your error was already fixed by 47 other teams. Success rate: 96%. One click to apply. The more teams use InariWatch, the faster everyone's errors get fixed.
Auto-Heal
Site goes down? InariWatch rolls back to the last working deploy in 30 seconds, then fixes the bug in the background. Total downtime: ~90 seconds. No human needed.
MCP Server
Your editor becomes the ops dashboard
InariWatch runs as an MCP server alongside your AI editor. Claude Code, Cursor, and Windsurf can query alerts, trigger fixes, and roll back deployments — without leaving your code.
{
"mcpServers": {
"inariwatch": {
"command": "inariwatch",
"args": ["serve-mcp"]
}
}
}Drop this in your project root. The server starts automatically — no daemon, no ports. Works the same in Claude Code, Cursor, and Windsurf.
5 action tools
get_root_cause
Deep AI analysis of an alert — pulls Sentry stack traces, Vercel build logs, and GitHub CI output in parallel.
trigger_fix
Full remediation pipeline: diagnose → read code → AI fix → self-review → push branch → wait CI → open PR.
rollback_vercel
Instantly roll back to the last successful production deployment. No CLI flags needed.
get_build_logs
Fetch Vercel build logs with automatic error extraction — ready to paste into a prompt.
silence_alert
Mark an alert as resolved from inside your editor once you've handled it.