Zelaxy

🌱 What is Zelaxy?

Zelaxy is a visual operating system for AI work. You compose workflows on a canvas out of blocks — AI agents, integrations, logic, loops, sub-workflows — connect them, and a real execution engine runs the graph with token-by-token LLM streaming.

It looks like a flow builder. Under the hood it's a concurrent, topological execution engine with parallel branches, loops, conditional routing, sub-workflows, built-in vector search (RAG), guardrails, and multi-provider AI orchestration — the kind of thing you'd otherwise stitch together from a dozen scripts and services.

…and growing every week. 🌾

🍃 Build workflows, not glue code. Drag three blocks onto the canvas, connect them, hit run.

🤖 Bring your own AI — 25+ providers, one canvas

Chain frontier labs, blazing-fast inference hosts, your own cloud, and local models — in a single pipeline. Pick a model from a dropdown; Zelaxy handles each provider's API, auth, streaming, and quirks.

🔌 Connect everything — 245+ integrations

Every logo below is a real block you drag onto the canvas. Auth, API calls, and pagination are handled for you.

📑 Table of Contents

• Why Zelaxy
• Features
• Build anything — example workflows
• How it works
• Blocks & integrations
• Quickstart
• Configuration
• Development
• Deployment
• Tech stack
• Contributing
• Sponsors
• License

🌸 Why Zelaxy

Most automation tools make you choose: great integrations or real AI or open-source & self-host. Zelaxy is built to give you all three on one canvas.

Zelaxy is right for you if you want to:

• ⚡ Automate complex AI processes without writing a custom backend
• 🔀 Chain multiple LLM providers (OpenAI, Claude, Gemini, Groq, DeepSeek, Grok, Mistral, Ollama…) in one pipeline
• 🔍 Visually debug — see exactly which block produced which output, and how long it took
• 📚 Build RAG pipelines with vector search built in — upload, embed, and query from any block
• 🪝 Trigger from Slack, GitHub, Gmail, Stripe, Telegram, webhooks, or cron
• 🏠 Self-host an open-source platform you can extend with your own blocks and tools

🚀 Features

🔬 Build anything

Six real, end-to-end workflows you can rebuild in minutes — each is a full walkthrough in the Guides →.

🔭 Deep Research Agent

Fans out to web + academic search in parallel, then an agent synthesizes a cited brief as structured JSON.

🗨️ Chat → 🔀 Parallel [ Tavily · ArXiv ] → 🤖 Agent · synthesize + cite → 📤 Response

🎧 Customer Support Agent

Classifies inbound email, drafts a knowledge-grounded reply, and escalates urgent cases — behind a human approval gate.

📧 Gmail → 🤖 Agent · classify → 🧭 Router → 🔴 urgent → Slack · 🟢 normal → Knowledge + draft + 🙋 Human → Gmail · ⚪ spam → stop

📚 RAG Knowledge Agent

Answers strictly from your documents via pgvector search, with a hallucination guardrail that blocks ungrounded replies.

🗨️ Chat → 🧠 Knowledge · vector search → 🤖 Agent · answer from passages → 🛡️ Guardrails · grounded? → 📤 Response

✍️ Multi-Agent Content Pipeline

A researcher → writer → editor hand-off turns one topic into publish-ready copy via three specialized agents.

📝 Topic → 🔎 Researcher · Tavily tool → ✍️ Writer · draft → 🪄 Editor · fact-check → 📤 Response

🎯 Lead Enrichment Agent

Turns a raw email into a scored, enriched CRM record — enrich, qualify with an agent, then branch on the score.

🪝 Webhook → 🔍 Hunter / Clay · enrich → 🤖 Agent · score fit → 🔀 Condition ≥ 70 → ✅ CRM write · ❌ drop

📅 Scheduled Report Agent

On a cron schedule, pulls metrics, has an agent write a narrative summary, and delivers it to Slack every morning.

⏰ Schedule · cron → 📊 API / Supabase / Sheets → 🤖 Agent · narrate → 💬 Slack

🛠 How it works

🧩 Blocks & integrations

Full, searchable reference at docs.zelaxy.in.

⚡ Quickstart

Open source. Self-hosted. No account required.

git clone https://github.com/manu14357/Zelaxy.git
cd Zelaxy
bun install
cp apps/zelaxy/.env.example apps/zelaxy/.env.local
# Edit .env.local — set DATABASE_URL, auth secrets, and at least one LLM API key
cd apps/zelaxy && bunx drizzle-kit migrate && cd ../..
bun run dev:full

App runs at http://localhost:3000, docs at http://localhost:3001.

Prefer hosted? Try zelaxy.in · Docs: docs.zelaxy.in

Requirements: Bun ≥ 1.2 · PostgreSQL 17 with pgvector · at least one LLM API key.

🔧 Configuration

LLM providers (add any): OPENAI_API_KEY, ANTHROPIC_API_KEY, GOOGLE_API_KEY, GROQ_API_KEY, DEEPSEEK_API_KEY, XAI_API_KEY, CEREBRAS_API_KEY, NVIDIA_API_KEY, MISTRAL_API_KEY, OPENROUTER_API_KEY

Local models: OLLAMA_URL (default http://localhost:11434), plus any OpenAI-compatible base URL (vLLM, LiteLLM, LM Studio).

💻 Development

bun run dev:full          # Full dev (app + websocket + docs)
bun run dev               # App only
bun run dev:sockets       # WebSocket server only
bun run build             # Build all
bun run test              # Run tests (~900 unit tests)
bun run lint              # Lint and auto-fix (Biome)
bun run type-check        # Type check (0 errors)

cd apps/zelaxy && bunx drizzle-kit studio    # Database GUI
cd apps/zelaxy && bunx drizzle-kit migrate   # Run migrations
bun run start:worker                          # Background job worker (requires REDIS_URL)

🌐 Deployment

The Next.js app deploys to Vercel. Two long-running services run on Railway:

Set NEXT_PUBLIC_SOCKET_URL and SOCKET_SERVER_URL on Vercel to your Railway socket URL. Background jobs use BullMQ backed by Redis (Upstash recommended) — set REDIS_URL on both Vercel and the Railway worker.

🧱 Tech stack

🤝 Contributing

Contributions are welcome — blocks, integrations, bug fixes, docs, all of it. 🌱

1. Fork the repository
2. Create a feature branch (git checkout -b feat/your-feature)
3. Make your changes
4. Run bun run test, bun run lint, and bun run type-check
5. Open a pull request

New here? Look for good first issue and read CONTRIBUTING.md.

💖 Sponsors

Zelaxy is free and open source. Sponsors keep it that way. 🌻

📄 License

MIT © 2025 Zelaxy — see LICENSE.