I was paying $600/mo for 5 SaaS tools. So I built one that does all of them — and open-sourced it.

#agents #ai #opensource #showdev

Last year I was running a 12-person team and paying for:

📋 Trello — task management ($12/user/mo)
📝 Notion — knowledge base ($10/user/mo)
⚡ Zapier — workflow automation ($30/mo)
📧 Gmail — email (free, but 5 tabs of context switching)
🤖 Some chatbot tool — AI assistant ($50/mo)

Total: ~$600/month. Five tools. Five bills. Five places where information went to die. 😩

I kept thinking: why can't one thing do all of this? 🤔

So I built HiveOps 🐝 — a self-hosted company operating system with AI agents that actually execute tasks, not just chat.

And I open-sourced it. 🎉

🛠 What HiveOps Actually Does

It's a full-stack Node.js app that gives your team:

📋 Task Management — priority levels, status tracking, assign to users or AI agents, comments, file attachments.

🤖 AI Agents — not chatbots. Real agents that execute tasks with retry logic, exponential backoff, and a dead letter queue. Assign a task to an agent and it runs it. ⚡

🔄 Workflow Automation — event-driven rules. Trigger → Conditions → Actions. Example: "When an urgent task is created, notify the department manager and auto-assign it to an agent."

📧 Real Email — inbox, sent, drafts, starred. IMAP polling for inbound. SMTP for outbound. AI drafts replies. Not a mock — real email. ✉️

💬 Real-Time Chat — Socket.IO with typing indicators, channels, direct agent chat.

📚 Knowledge Base — full-text search, categories, CRUD.

⏰ Scheduler — DB-persisted cron. Interval, daily, weekly. Auto-creates tasks and triggers agents.

🏗 The Tech Stack

🖥️ Backend:    Node.js + Express
🎨 Frontend:   Next.js 14 (App Router) + Tailwind
💾 Database:   SQLite (dev) or PostgreSQL (production)
⚡ Real-time:  Socket.IO
🔐 Auth:       JWT + bcrypt
🧠 AI:         Any OpenAI-compatible API
📧 Email:      Nodemailer + ImapFlow
🚀 Deploy:     docker-compose up

The database adapter is the part I'm most proud of. Same API, two backends:

// This works on SQLite AND PostgreSQL — zero changes needed ✨
const user = await db.prepare('SELECT * FROM users WHERE id = ?').get(id);
const tasks = await db.prepare('SELECT * FROM tasks WHERE status = ?').all('pending');
await db.prepare('INSERT INTO users (...) VALUES (...)').run(...);

SQLite for local dev (no Docker needed 🙌). PostgreSQL for production (one env var). Auto-converts ? placeholders to $1, $2. Auto-handles boolean differences. Auto-coerces bigint strings from COUNT queries. 🔧

🤖 The AI Agent System (This Is The Interesting Part)

Most "AI agent" projects are chatbots with extra steps. HiveOps agents are different:

// 🔄 Agent auto-execution loop:
// 1. Pick pending tasks assigned to active agents
// 2. Execute via LLM with the agent's system prompt
// 3. On failure -> retry with exponential backoff (10s, 20s, 40s)
// 4. After 3 retries -> dead letter queue + notify task creator 💀

Agents can also delegate to each other 🤝:

// Agent A (marketing) delegates to Agent B (engineering)
await agentDelegate(marketingAgentId, engineeringAgentId,
  "We need a landing page for the Q2 campaign", taskId);

And the workflow engine ties it all together 🔗:

// When a task is created with priority "urgent":
// -> Check: is priority === "urgent"? ⚠️
// -> Action 1: notify the department manager 📢
// -> Action 2: auto-assign to the ops agent 🤖
// -> Action 3: send a message to the #urgent channel 💬

No Zapier. No monthly fee. Just rules that run. ✅

🚀 How to Run It

git clone https://github.com/mamoor123/hiveops.git && cd hiveops
cp .env.example .env
openssl rand -base64 32  # paste into JWT_SECRET 🔑
docker-compose up --build

Open http://localhost:3000. That's it. 🎯

Or if you don't want Docker — SQLite mode works out of the box:

cd server && npm install && npm run migrate
JWT_SECRET=your-secret npm run dev

🧠 What I Learned Building This

1️⃣ Dual-database adapters are worth the effort. Supporting SQLite + PostgreSQL from day one means devs can contribute without Docker, and you deploy with Postgres for real workloads. The adapter is ~250 lines and handles all the quirks (bigint strings, boolean differences, placeholder syntax). 🔧

2️⃣ AI agents need execution infrastructure, not just prompts. The prompt is 10% of the work. The retry logic, error handling, dead letter queues, concurrency control — that's the other 90%. 💪

3️⃣ Real-time is a superpower for internal tools. Socket.IO notifications feel alive compared to polling. Users notice the difference. ⚡

4️⃣ Open source is a force multiplier. I built this for myself, but open-sourcing it means I get bug reports, feature requests, and contributors for free. 🌍