BlameBot — AI-Powered Incident Intelligence

Vercel Zero to Agent Hackathon submission · Solo · Deployed on Vercel

BlameBot is an autonomous on-call agent that closes the loop from deploy failure to resolved incident — without waking up your whole team at 3 AM. It ingests alerts from Vercel, Sentry, and UptimeRobot, uses AI to explain what broke and who owns it, pages the right person via Slack, and can roll back the deployment automatically.

Live Demo → · GitHub →

The Problem

Every on-call rotation has the same failure loop:

1. Alert fires at 2 AM
2. Engineer scrambles to find the right Slack thread, the right commit, the right person
3. 40 minutes later: "oh it was that one deploy"
4. Team manually writes a postmortem (or doesn't)

BlameBot collapses that 40-minute scramble into a 10-second AI summary with one-click rollback.

How It Works

Webhook (Vercel / Sentry / UptimeRobot)
        │
        ▼
┌───────────────────────────────────────────────────────┐
│                    Incident Pipeline                   │
│                                                       │
│  Normalize → Enrich (GitHub) → Resolve Owners         │
│       → Decide Severity → Find Similar Incidents      │
│       → AI Report (Claude Sonnet) → Persist (Redis)   │
│       → Post to Slack → Schedule Escalation           │
└───────────────────────────────────────────────────────┘
        │
        ▼
  Slack alert with AI summary, blame, and action buttons
  (Acknowledge / Rollback / Draft Hotfix PR)
        │
        ▼
  Dashboard + Analytics + Postmortem

Pipeline Steps

Features

Live Dashboard

• Active incident count with severity breakdown
• Real-time incident feed
• Integration health status (Vercel, Sentry, UptimeRobot)
• Top offenders — developers causing the most incidents
• Activity feed across all ongoing incidents

AI-Powered Incident Reports

• Root cause analysis written in plain English
• Blast radius assessment (affected services, users, regions)
• Recommended action (rollback, hotfix, monitor)
• Automatic similarity detection against recent incidents
• One-click postmortem generation

Slack-First Response

• Rich Block Kit incident cards with all context in one message
• Acknowledge / Rollback / Draft Hotfix PR buttons
• Thread-based timeline updates
• Escalation alerts when incidents go unacknowledged

Analytics

• Incident trends over time charted by severity
• MTTR (Mean Time to Resolution) tracking
• Top problematic files and services
• Developer reliability scores and leaderboard
• Risk heatmap — incidents by day of week and hour
• Service health scores

Configuration Panel

• Owner Rules — file glob → Slack user/group mapping
• Severity Rules — pattern-based auto-classification
• Escalation Policy — per-severity delay and escalation contacts
• Blackout Windows — timezone-aware quiet hours with fallback contact
• Autopilot — auto-rollback thresholds per severity level
• Natural Language Config — update any setting in plain English via AI

Autopilot Mode

When enabled, BlameBot can automatically roll back a failing deployment via the Vercel API without any human intervention. Configurable per severity level with a confirmation grace window.

Tech Stack

Setup

Prerequisites

• Node.js 18+
• pnpm
• Upstash Redis instance
• Vercel project
• Slack app (bot token + signing secret)
• GitHub personal access token

1. Clone and install

git clone https://github.com/VusalHuseynov/blamebot
cd blamebot
pnpm install

2. Configure environment variables

Create .env.local in the project root:

# ── App ────────────────────────────────────────────────
APP_URL=http://localhost:3000
NEXT_PUBLIC_APP_URL=http://localhost:3000

# ── Upstash Redis ───────────────────────────────────────
# Create a database at https://console.upstash.com
UPSTASH_REDIS_REST_URL=https://your-db.upstash.io
UPSTASH_REDIS_REST_TOKEN=your_token
KV_REST_API_URL=https://your-db.upstash.io
KV_REST_API_TOKEN=your_token

# ── GitHub ──────────────────────────────────────────────
# PAT with repo + read:org scopes
GITHUB_TOKEN=ghp_xxxxxxxxxxxxxxxxxxxx
GITHUB_OWNER=your-org-or-username
GITHUB_REPO=your-repo-name
# For GitHub OAuth (optional)
GITHUB_CLIENT_ID=Iv1.xxxxxxxxxxxx
GITHUB_CLIENT_SECRET=your_secret

# ── Slack ───────────────────────────────────────────────
# Create an app at https://api.slack.com/apps
SLACK_BOT_TOKEN=xoxb-xxxxxxxxxxxx
SLACK_SIGNING_SECRET=your_signing_secret
SLACK_INCIDENTS_CHANNEL_ID=C0XXXXXXXXX
# For Slack OAuth (optional)
SLACK_CLIENT_ID=your_client_id
SLACK_CLIENT_SECRET=your_client_secret

# ── Vercel ──────────────────────────────────────────────
# API token from https://vercel.com/account/tokens
VERCEL_API_TOKEN=your_vercel_token
VERCEL_PROJECT_ID=prj_xxxxxxxxxxxx
# VERCEL_TEAM_ID only needed for team projects
VERCEL_TEAM_ID=team_xxxxxxxxxxxx

# ── Webhook secrets ─────────────────────────────────────
# Generate with: openssl rand -hex 32
VERCEL_WEBHOOK_SECRET=your_secret
SENTRY_WEBHOOK_SECRET=your_secret
UPTIMEROBOT_WEBHOOK_SECRET=your_secret

# ── Admin ───────────────────────────────────────────────
ADMIN_TOKEN=your_admin_token
ADMIN_COOKIE_SECRET=your_cookie_secret

# ── Cron ────────────────────────────────────────────────
CRON_SECRET=your_cron_secret

3. Run locally

pnpm dev

Open http://localhost:3000.

4. Seed demo data (optional)

curl -X POST http://localhost:3000/api/seed

This populates Redis with sample incidents, timeline events, and developer scores so the dashboard is populated immediately.

Webhook Setup

Point these services at your deployed URL:

Vercel

Settings → Webhooks → Add webhook

URL: https://your-app.vercel.app/api/webhooks/vercel
Events: deployment.error, deployment.canceled

Sentry

Project Settings → Developer Settings → Internal Integration → Webhooks

URL: https://your-app.vercel.app/api/webhooks/sentry
Events: issue (created, resolved)

UptimeRobot

My Settings → Alert Contacts → Add Alert Contact

Type: Web-hook
URL: https://your-app.vercel.app/api/webhooks/uptime

All webhook endpoints validate HMAC signatures. Set the corresponding *_WEBHOOK_SECRET env vars to match what each service sends.

Slack App Setup

1. Go to api.slack.com/apps → Create New App → From manifest

2. Use this manifest:

display_information:
  name: BlameBot
  description: AI-powered incident intelligence
features:
  bot_user:
    display_name: BlameBot
    always_online: true
  slash_commands:
    - command: /incident
      url: https://your-app.vercel.app/api/slack/commands
      description: Query BlameBot about incidents
  interactivity:
    is_enabled: true
    request_url: https://your-app.vercel.app/api/slack/interactivity
oauth_config:
  scopes:
    bot:
      - chat:write
      - channels:history
      - channels:read
      - commands
settings:
  event_subscriptions:
    request_url: https://your-app.vercel.app/api/slack/events
    bot_events:
      - message.channels

1. Install to workspace, copy Bot User OAuth Token → SLACK_BOT_TOKEN
2. Copy Signing Secret → SLACK_SIGNING_SECRET

Deployment

BlameBot is built for Vercel. Push to your repo and connect the project.

The included vercel.json configures a daily escalation cron:

{
  "crons": [
    {
      "path": "/api/cron/escalations",
      "schedule": "0 0 * * *"
    }
  ]
}

Set all environment variables in the Vercel dashboard under Settings → Environment Variables.

API Reference

Project Structure

blamebot/
├── app/
│   ├── (pages)/          # Next.js App Router pages
│   │   ├── dashboard/
│   │   ├── incidents/
│   │   ├── analytics/
│   │   └── config/
│   └── api/              # API routes
│       ├── webhooks/     # Webhook receivers
│       ├── incidents/    # Incident CRUD
│       ├── config/       # Configuration endpoints
│       ├── slack/        # Slack event/command/interactivity
│       ├── analytics/    # Analytics aggregation
│       └── cron/         # Scheduled jobs
├── lib/
│   ├── pipeline/         # Core incident processing
│   │   ├── index.ts      # Orchestrator
│   │   ├── normalize.ts  # Webhook normalization
│   │   ├── enrich.ts     # GitHub enrichment
│   │   ├── severity.ts   # Severity classification
│   │   ├── history.ts    # Similar incident search
│   │   ├── ai.ts         # Claude AI integration
│   │   ├── devscore.ts   # Developer scoring
│   │   └── actions.ts    # Rollback / hotfix actions
│   ├── schemas.ts        # Zod schemas
│   ├── types.ts          # TypeScript types
│   └── redis.ts          # Redis client
├── components/           # React components
└── vercel.json           # Cron job config

Built for Vercel Zero to Agent Hackathon

This project was built as a solo submission for the Vercel Zero to Agent Hackathon (April 24 – May 3, 2025).

Why BlameBot fits the "Zero to Agent" theme:

• It is a genuine autonomous agent, not a chatbot. It receives a signal, reasons about it, takes action, and reports back — all without human involvement.
• The agent loop spans multiple tools: GitHub for code context, Vercel API for rollback actions, Slack for communication, and Claude for reasoning.
• Autopilot mode lets it close the full loop: detect → analyze → act → report — zero human steps required.

License

MIT