AI agents that visit
your workspace

A sovereign workspace where AI agents get real tools — terminals, browsers, voice, design software — on any machine you control.

View on GitHub Download for Windows

They don't run in a chat window. They show up in a terminal and write your backend. They open a browser and automate your workflows. They open design tools and build your interfaces.

The output isn't a message — it's files, running code, deployed services.

Mahana Workspace Tile 1: Claude Tile 2: Grok Writing a REST API Automating browser tests [terminal + editor] [browser + screenshots] Tile 3: Kimi Tile 4: GPT Designing the UI Reviewing PR #47 [design tool + voice] [terminal + git]

What you get

Provider-Agnostic

Claude, Grok, Kimi, GPT, Gemini, DeepSeek, local Ollama. Any model in any tile. Switch mid-task.

Real Tools

Full terminal, isolated browser, voice interface, 2150+ HTTP API routes. Not sandboxed — real.

Sovereign

Runs on your hardware. Your data stays yours. E2E encrypted relay between devices. No cloud required.

Multi-Device

Laptop, server, Pi, VM — every machine becomes a node. Control any device from any device.

Session Memory

Neuropacket compression (100:1). Agents remember context across sessions. RAG search across everything.

Open Source

MIT licensed. The workspace, tools, MCP servers, sovereign infrastructure — all free.

The Creation Loop

Spawn an agent
↓
Agent opens a tool (terminal, browser, voice)
↓
Agent creates something real (code, UI, data)
↓
Output becomes a new capability
↓
Next agent builds on it
↻

Free and Open Source

The workspace is the node. It's free because the value is in what visits it.

git clone https://github.com/itonsberg/mahana && pnpm dev Click to copy