Agentix

Open source visual builder for AI agents. No code. No token middleman. Runs local with Ollama (free) or with the Anthropic API (your key, your tokens). The builder is yours forever.

---

Why Agentix?

Every "AI agent platform" out there (Zapier, Make, n8n cloud, Relevance, etc.) charges you a fat margin on top of the tokens you'd pay the LLM provider directly. They host your logic on their servers and lock you in.

Agentix flips that. The builder is a single HTML file that runs entirely in your browser. You bring your own LLM — either Ollama running locally (totally free, your hardware) or your own Anthropic API key. We never see anything, we never run a server, we never lock you in.

The product is dead simple:

• Builder → free, MIT, runs locally, forever
• Marketplace → buy and sell pre-built agents (coming soon)
• Cloud → optional hosted execution if you want webhooks/cron (later)

---

Getting started

You don't need to install anything to use the builder. Seriously.

git clone https://github.com/YOUR_USERNAME/agentix.git
cd agentix
open builder.html   # or just double-click it

Or even simpler: download builder.html and open it in any modern browser.

Setting up Ollama (recommended — free, local, no API costs)

Ollama runs LLMs locally on your machine. Free forever, your hardware does the work.

1. Install Ollama → download from ollama.com (Mac, Linux, Windows)

2. Pull a model (in your terminal):

ollama pull llama3.2:3b      # 2.5GB · recommended for 16GB RAM Macs
# or
ollama pull qwen2.5:7b       # 5GB · smarter, needs more RAM
ollama pull llama3.1:8b      # 5GB · also good
ollama pull phi3.5           # 2.8GB · fast and capable

3. Enable CORS so the browser can talk to Ollama (one-time setup):

On macOS:

launchctl setenv OLLAMA_ORIGINS "*"

Then quit and restart the Ollama app from the menu bar.

On Linux:

# Edit the systemd service
systemctl edit ollama.service
# Add under [Service]:
Environment="OLLAMA_ORIGINS=*"
# Then:
systemctl daemon-reload && systemctl restart ollama

On Windows: set the OLLAMA_ORIGINS environment variable to * in System Properties → Environment Variables, then restart Ollama.

4. Open the builder, click ⚙ Provider, select Ollama, pick your model, and you're set.

Without OLLAMA_ORIGINS=* the browser will block the request with a CORS error. This is the most common gotcha.

Setting up Anthropic API (optional — paid per token)

If you want top-tier quality (Sonnet/Opus) for production agents:

1. Go to console.anthropic.com, create a key, add a few dollars of credit
2. Open the builder, click ⚙ Provider, select "Anthropic API", paste the key
3. Pick a model and run

Note: an Anthropic Pro subscription ($20/mo for claude.ai) does not include API access. They're separate billings.

First agent in 60 seconds

1. Drag a Trigger node onto the canvas
2. Drag a Claude LLM node next to it
3. Drag an Output node after that
4. Connect them: click and drag from the right port of one to the left port of the next
5. Click each node and fill in the configuration
6. Hit ▶ Run

That's it. You just built your first agent.

---

Node types (v0.1.0)

Node	What it does
Trigger	The starting point. Holds the initial input text.
Claude LLM	Sends a system + user prompt to Claude. Use {{input}} to reference the previous node's output.
HTTP Request	Call any REST API. GET / POST / PUT / DELETE.
Condition	If output contains X, branch to one of two values.
Output	Marks the final result of the agent.

More nodes coming: Slack, Email, Database, Webhook, Schedule, Loop, Parser.

---

How it works under the hood

┌───────────────────┐
│   builder.html    │  ← single file, no build, no backend
│  (runs in browser)│
└─────────┬─────────┘
          │
          ├──────────────────┐
          │                  │
          ↓                  ↓
┌──────────────────┐  ┌──────────────────┐
│  Ollama (local)  │  │  Anthropic API   │
│ localhost:11434  │  │ api.anthropic.com│
│   (free, your    │  │  (paid, your     │
│    hardware)     │  │     key)         │
└──────────────────┘  └──────────────────┘

That's the whole architecture. No database. No servers. No accounts. No telemetry.

When you "save" an agent, it exports as a .agentix.json file. When you "load" one, it reads that file. Your work is yours.

---

Roadmap

v0.1.0 — MVP (you are here)

• Visual canvas with drag-drop nodes
• 5 node types (Trigger, Claude LLM, HTTP, Condition, Output)
• Live execution in browser
• Import / export as JSON
• Local API key storage

v0.2.0 — Polish

• Multiple inputs per node
• Better edge routing (no crossings)
• Pan and zoom canvas
• Template gallery (10 starter agents)
• Dark / light themes

v0.3.0 — Marketplace

• Agent gallery site
• Stripe checkout for paid agents
• Creator profiles
• Rating & reviews
• 70/30 split (creator/platform)

v0.4.0 — Cloud (optional, paid)

• Hosted agent execution
• Webhook triggers
• Cron schedules
• Team workspaces

---

Marketplace economics (the business model)

Agentix the builder is and always will be free. The business runs on the marketplace:

Creator builds an agent  →  uploads to marketplace
                         ↓
Customer buys it for $39 →  pays via Stripe
                         ↓
                    Stripe takes 2.9%
                    Agentix takes 30%
                    Creator gets ~67%

Why this works:

• Zero marginal cost (no tokens to pay — buyers use their own keys)
• Network effect (more creators → more agents → more buyers → more creators)
• Aligned incentives (we only make money when creators do)

---

Contributing

Pull requests welcome. The whole thing is one HTML file (builder.html) for v0.1.0 — that's intentional. Easy to read, easy to fork, easy to modify.

Open an issue if you have ideas for new node types or want to report a bug.

---

License

MIT — do whatever you want with it.

---

Built by

Miguel — Mar del Plata, Argentina 🇦🇷

If you're using Agentix in production, drop me a line. I want to hear about it.