GitHub - abhay-codes07/operon-ai: The control plane for autonomous web operations. Durable, observable AI agents running real browser workflows in production.

 ██████╗ ██████╗ ███████╗██████╗  ██████╗ ███╗   ██╗
██╔═══██╗██╔══██╗██╔════╝██╔══██╗██╔═══██╗████╗  ██║
██║   ██║██████╔╝█████╗  ██████╔╝██║   ██║██╔██╗ ██║
██║   ██║██╔═══╝ ██╔══╝  ██╔══██╗██║   ██║██║╚██╗██║
╚██████╔╝██║     ███████╗██║  ██║╚██████╔╝██║ ╚████║
 ╚═════╝ ╚═╝     ╚══════╝╚═╝  ╚═╝ ╚═════╝ ╚═╝  ╚═══╝

The web is now programmable.

Operon is an autonomous web agent platform built on TinyFish AI. Deploy fleets of parallel agents that watch, extract, and act across any website — with full observability, intelligence, and control built in.

Live Demo · Dashboard · GitHub

Why TinyFish beats everything else

	Operon / TinyFish	OpenAI Operator	Claude Computer Use	Browser Use
Mind2Web Accuracy	89.9%	61.3%	56.3%	30.0%
Easy → Hard Degradation	−15.6 pts	−39 pts	−45 pts	−58 pts
Parallel Agents	1,000	1	1	1
Cost / step	$0.015	~$0.08	~$0.12	varies

TinyFish holds accuracy on compound multi-step tasks where every other tool collapses.

What Operon Does That Nothing Else Can

SnapBuy — Autonomous Deal Sniper

"Set it. Forget it. Own it."

Every other price tool sends you an alert. SnapBuy completes the purchase. Set a trigger condition in plain English — "price drops below $899", "GPU back in stock", "visa appointment slot opens". A TinyFish agent watches the URL. The instant the condition is met, it navigates, fills your saved payment details, and finishes checkout. Order confirmation in your inbox before you even see the notification.

trigger: "RTX 5090 back in stock at Best Buy"
action:  "add to cart, apply coupons, checkout with saved card"
result:  Order #BB-2847291 confirmed ✓

Heartbeat — Synthetic User Monitoring

"Does your product still work? Find out before your users do."

Define your product's critical flows in plain English. Every 15 minutes, a TinyFish agent acts as a real user going through them: sign up, checkout, create a project, invite a teammate. If anything breaks — a button doesn't render, a form errors out, a page is blank — you get an alert with a screenshot of exactly what failed and at which step. Not HTTP 200 checks. Visual, human-accuracy, UI-level monitoring.

journey: "New User Signup → Onboarding → First Agent Created"
status:  Step 3 FAILED — "Stripe form did not render within 8s"
alert:   Sent 19 minutes ago with screenshot evidence

PriceWatch — Price Drop Intelligence

"Never pay full price again."

Add any product from any retailer. Set your target price. TinyFish checks it on your schedule — hourly, every 6 hours, daily. The moment the price drops below your threshold, you get an instant email + Slack notification with a direct link. Sparkline price history so you can see the trend. Lowest-ever price shown so you know if it's actually a deal.

watching: Sony WH-1000XM5 on Amazon  ($279 → alert at $249)
history:  $349 → $329 → $299 → $279 → ... → 🔔 $249

EthicsWatch — Real-Time ESG & Ethics Intelligence

"Know when organizations change their commitments. Before the market does."

Hedge funds, policymakers, NGOs, and compliance teams monitor organizations for changes in ESG standards, corporate social responsibility commitments, new regulations, governance policies, and supply chain ethics. TinyFish agents scrape corporate CSR pages, SEC EDGAR filings, regulatory body announcements, and UN SDG databases. Claude analyzes whether a change is materially significant — you get a briefing, not a raw diff.

monitor: SEC climate disclosure rules
alert:   CRITICAL — "SEC finalized Scope 1+2 GHG reporting for all large filers"
impact:  "Every public company must update 10-K filings. Non-compliance = material misstatement."

Swarm Orchestration — Fleet at Scale

One command launches parallel agents across multiple URLs simultaneously. Live status grid. Results aggregate automatically when all agents report back.

POST /api/internal/swarm/launch
{ "targetUrls": ["amazon.com", "ebay.com", "walmart.com", "bestbuy.com"] }
# → 4 agents running in parallel right now

Sentinel Watchlist — Semantic Change Detection

Deploy always-on agents monitoring specific URLs on a schedule. Unlike DOM diffing, Claude determines if a change is semantically meaningful and generates an intelligence briefing: "Amazon renamed their Prime tier and added a new upsell step to checkout — your selector needs updating."

Agent DNA Transfer — Behavioral Fingerprinting

After each successful execution, Operon extracts a behavioral fingerprint — exact action sequences, selector patterns, timing confidence — and makes it transferable to new agents. Near-instant first-run success. No warmup.

Recon — Autonomous Security Scanner

Six TinyFish agents simultaneously scan a target domain: admin panel exposure, sensitive file disclosure (.env, .git/config), login error enumeration, directory listings, unauthenticated API endpoints, missing security headers. Severity-rated findings in minutes.

Pulse — Competitive Intelligence Engine

Monitor any competitor domain with 3 parallel agents: pricing tiers, hiring signals (8 ML engineers hired = new AI feature in 90 days), feature page changes. Claude synthesizes a battle card: their weaknesses, your advantages, a recommended counter.

DataMesh — Web-to-JSON Extraction API

Public REST API. Define a schema. Get structured data from any URL.

curl -X POST https://your-operon.vercel.app/api/v1/extract \
  -H "x-api-key: op_live_xxxx" \
  -d '{"url":"amazon.com/product/...", "schema":{"price":"number","title":"string"}}'
# → {"data":{"price":979.99,"title":"Apple iPhone 16 Pro"}, "confidence":0.96}

Results Hub — Unified Output Dashboard

Every execution result in one place — prices, jobs, extracted data, structured output. Smart preview auto-detects result type. Auto-refreshes every 3 seconds while agents are active.

More Built-In

Feature	What it does
Real-Time SSE	Live execution events stream to browser — no WebSocket server needed
NL Workflow Builder	Describe a task in English → Claude builds the full workflow definition
Agent Leaderboard	Composite reliability ranking: success rate, retry penalty, latency
Operon Shield	Prompt injection defense — injections scoring ≥70 blocked before execution
Self-Healing Selectors	Semantic fallback: tries `data-testid`, `aria-label`, fuzzy similarity
Failure RCA	Auto-classifies failures: `SELECTOR_DRIFT`, `AUTH_ISSUE`, `PAGE_LOAD_TIMEOUT`
Co-Pilot	Human-in-the-loop — agent pauses for operator approval when confidence drops
Mission Control	Fleet-level incident detection + automated runbook execution
FinOps	Per-run cost tracking, anomaly detection, per-workflow budget enforcement
Compliance Passport	Full audit trail + plain-English compliance summary via Claude
Canary Releases	Progressive rollout with automatic rollback on failure threshold
Screenshot Gallery	Visual proof-of-work — every step captured with keyboard navigation

Architecture

╔══════════════════════════════════════════════════════════╗
║                     Next.js 14 App                        ║
║  ┌─────────────┐   ┌──────────────┐   ┌──────────────┐   ║
║  │ App Router  │   │  90+ API     │   │  NextAuth    │   ║
║  │ RSC + CC    │   │  Routes      │   │  RBAC        │   ║
║  └─────────────┘   └──────────────┘   └──────────────┘   ║
╚═════════════════════════════╦════════════════════════════╝
                              ║ BullMQ jobs
              ╔═══════════════▼═════════════════╗
              ║       Execution Worker           ║
              ║  tinyfish-execution-runner.ts     ║
              ║  ┌─────────────────────────────┐ ║
              ║  │   TinyFish Web Agent API    │ ║
              ║  │   89.9% Mind2Web Accuracy   │ ║
              ║  │   SSE streaming · 1000 agt  │ ║
              ║  └─────────────────────────────┘ ║
              ║  Shield · DNA · FinOps · Alerts  ║
              ╚═══════════════╦═════════════════╝
                              ║
              ╔═══════════════▼═════════════════╗
              ║     PostgreSQL via Prisma        ║
              ║     50+ models · migrations      ║
              ╚═════════════════════════════════╝

Patterns:

Server Components for all data fetching — zero client-side API overhead
BullMQ Worker as separate process (npm run worker:dev)
SSE at /api/internal/executions/[id]/sse — no WebSocket server
Repository pattern: services → repositories → Prisma

Quick Start

Prerequisites

Node.js 20+
Docker
TinyFish API key — tinyfish.ai
Anthropic API key — console.anthropic.com (optional)

1. Clone & install

git clone https://github.com/abhay-codes07/operon-ai.git
cd operon-ai
npm install

2. Configure environment

cp .env.example .env
# Required: DATABASE_URL, REDIS_URL, TINYFISH_API_KEY, NEXTAUTH_SECRET

3. Start infrastructure

docker-compose up -d   # PostgreSQL :5433, Redis :6379

4. Set up database

npx prisma migrate deploy
npx prisma db seed

5. Run

npm run dev          # Terminal 1 — Next.js on :3000
npm run worker:dev   # Terminal 2 — BullMQ worker (required for agent execution)

Deploy to Vercel

1. Push to GitHub (already done)

git push origin main

2. Create Vercel project

npm i -g vercel
vercel --prod
# Or: go to vercel.com → New Project → Import from GitHub

3. Set environment variables in Vercel dashboard

TINYFISH_API_KEY=...
DATABASE_URL=postgresql://...         # Use Neon, Supabase, or Railway
REDIS_URL=redis://...                 # Use Upstash Redis
NEXTAUTH_SECRET=...                   # openssl rand -base64 32
NEXTAUTH_URL=https://your-app.vercel.app
ANTHROPIC_API_KEY=...                 # optional

4. Deploy worker separately

The BullMQ worker cannot run on Vercel (serverless). Deploy it to:

Railway: railway up — runs the worker process persistently
Render: create a Background Worker service pointing to worker/index.ts
Fly.io: fly deploy — Dockerfile included

5. Database options (managed PostgreSQL)

Provider	Free tier	Notes
Neon	✅ 3GB	Serverless, best for Vercel
Supabase	✅ 500MB	Includes auth + storage
Railway	✅ 1GB	Also hosts the worker

6. Redis options

Provider	Free tier	Notes
Upstash	✅ 10K req/day	Serverless Redis, perfect for Vercel
Railway Redis	✅ 1GB	Co-locate with worker

Environment Variables

Variable	Required	Description
`TINYFISH_API_KEY`	✅	TinyFish Web Agent API key
`DATABASE_URL`	✅	PostgreSQL connection (port 5433 locally)
`REDIS_URL`	✅	Redis connection string
`NEXTAUTH_SECRET`	✅	32-byte random secret (`openssl rand -base64 32`)
`NEXTAUTH_URL`	✅ prod	Full URL of your deployment
`SMTP_HOST` + `ALERT_EMAIL_TO`	optional	Email alerts
`SLACK_WEBHOOK_URL`	optional	Slack notifications
`DATAMESH_API_KEY`	optional	Public DataMesh API key validation

Tech Stack

Layer	Technology
Framework	Next.js 14 (App Router)
Language	TypeScript 5
Styling	Tailwind CSS
Database	PostgreSQL + Prisma ORM
Queue	BullMQ + Redis
Auth	NextAuth.js v4
Web Agents	TinyFish Web Agent API
Billing	Stripe
Email	Nodemailer (SMTP)
Deployment	Vercel + Railway/Render (worker)

Project Structure

src/
├── app/
│   ├── dashboard/
│   │   ├── snapbuy/        # Autonomous deal sniper
│   │   ├── heartbeat/      # Synthetic user monitoring
│   │   ├── pricewatch/     # Price drop alerts
│   │   ├── ethicswatch/    # ESG & ethics intelligence
│   │   ├── recon/          # Security surface scanner
│   │   ├── pulse/          # Competitive intelligence
│   │   ├── datamesh/       # Web-to-JSON API playground
│   │   ├── results/        # Unified results hub
│   │   ├── swarm/          # Swarm orchestrator
│   │   ├── sentinels/      # Sentinel watchlist
│   │   ├── agents/         # Agent registry + leaderboard
│   │   ├── activity/       # Execution timeline + detail
│   │   ├── workflows/      # Workflow builder
│   │   └── ...             # 15+ more pages
│   ├── api/
│   │   ├── internal/       # 90+ authenticated routes
│   │   └── v1/             # Public API (DataMesh)
│   └── page.tsx            # Landing page
├── components/dashboard/
│   ├── snapbuy/            # SnapbuyDashboard
│   ├── heartbeat/          # HeartbeatDashboard
│   ├── pricewatch/         # PricewatchDashboard
│   ├── ethicswatch/        # EthicswatchDashboard
│   ├── recon/              # ReconLauncher
│   ├── pulse/              # PulseDashboard
│   ├── datamesh/           # DataMeshPlayground
│   └── ...
├── server/
│   ├── services/executions/  # tinyfish-execution-runner.ts
│   ├── services/sentinels/
│   ├── services/notifications/
│   └── integrations/tinyfish/
└── config/

prisma/schema.prisma          # 50+ models
worker/                       # BullMQ worker (separate process)

6-Minute Demo Script

Time	Scene
0:00–0:25	Landing page — animated agent network, benchmark chart
0:25–1:45	SnapBuy — RTX 5090 TRIGGERED card, price progress bars, create a snipe
1:45–3:00	Heartbeat — DEGRADED checkout flow, EKG chart, failure screenshot
3:00–3:45	PriceWatch — sparkline charts, ALERT_SENT amber card, price history
3:45–4:30	EthicsWatch — SEC CRITICAL alert, Apple CSR change, battle card
4:30–4:50	Swarm — launch 4 agents simultaneously, status grid
4:50–5:10	DataMesh — schema builder, extract JSON, copy curl
5:10–5:30	Results Hub + Agent Leaderboard
5:30–6:00	Landing page close — "The web is now programmable."

Built on TinyFish AI · TinyFish Hackathon 2026

"The web moves faster than you do. Now you don't have to."

Name		Name	Last commit message	Last commit date
Latest commit History 644 Commits
.claude		.claude
.logs		.logs
docs		docs
prisma		prisma
src		src
.dockerignore		.dockerignore
.editorconfig		.editorconfig
.env.example		.env.example
.eslintrc.json		.eslintrc.json
.gitignore		.gitignore
.prettierignore		.prettierignore
.prettierrc.json		.prettierrc.json
Dockerfile		Dockerfile
README.md		README.md
check-execution.js		check-execution.js
docker-compose.yml		docker-compose.yml
middleware.ts		middleware.ts
next.config.mjs		next.config.mjs
package-lock.json		package-lock.json
package.json		package.json
postcss.config.mjs		postcss.config.mjs
setup-tinyfish-key.mts		setup-tinyfish-key.mts
tailwind.config.ts		tailwind.config.ts
tsconfig.json		tsconfig.json
vercel.json		vercel.json

Folders and files

Latest commit

History

Repository files navigation

The web is now programmable.

Why TinyFish beats everything else

What Operon Does That Nothing Else Can

SnapBuy — Autonomous Deal Sniper

Heartbeat — Synthetic User Monitoring

PriceWatch — Price Drop Intelligence

EthicsWatch — Real-Time ESG & Ethics Intelligence

Swarm Orchestration — Fleet at Scale

Sentinel Watchlist — Semantic Change Detection

Agent DNA Transfer — Behavioral Fingerprinting

Recon — Autonomous Security Scanner

Pulse — Competitive Intelligence Engine

DataMesh — Web-to-JSON Extraction API

Results Hub — Unified Output Dashboard

More Built-In

Architecture

Quick Start

Prerequisites

1. Clone & install

2. Configure environment

3. Start infrastructure

4. Set up database

5. Run

Deploy to Vercel

1. Push to GitHub (already done)

2. Create Vercel project

3. Set environment variables in Vercel dashboard

4. Deploy worker separately

5. Database options (managed PostgreSQL)

6. Redis options

Environment Variables

Tech Stack

Project Structure

6-Minute Demo Script

About

Resources

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages