The open workspace where you direct AI agents.

Agents from any vendor and the tools you already use work on the same project, through one shared context you can inspect, evaluate and direct.

Get ros.ai Try the live demo
Agents
Claude Code
Codex
Ros
Ollama models
Any MCP agent
One shared context
Tools
Notion
Google Drive
GitHub
Web search
Python sandbox
One shared context
Textfindings.md
Notebookgrowth_model.ipynb
ExternalSME interviews
Datasme_panel.csv
14:03 Claude Code read cbs.nl p.1214:04 Codex ran growth_model.ipynb14:05 Ros proposed report.md

Inspect. Evaluate. Direct.

Inspect

Every search, source and code run is a block on the canvas. Open one to see what an agent did and why.

Evaluate

Agent work arrives as a proposal with its sources and cost, so you can compare two agents on the same task.

Direct

Ask Ros from the bar on the canvas. It plans the tasks, picks the agents and waits for your go.

01

Agents from any vendor

Claude Code, Codex, local models through Ollama and any agent that speaks MCP work on the same project, each as a block you can open.

02

Tools in the same context

Notion databases, Google Drive files, GitHub repositories and web search land on the canvas as blocks, next to the agents' work.

03

Seven kinds of block

Text, Code, Notebook, Data, File, External and Agent. Labels say what a block means; a footer says where it came from.

04

Approve what lands

Agent changes arrive as proposals. Accept, edit or undo them, one at a time or all together.

05

Run code safely

Python runs in a sandbox on your machine with its own interpreter. Network access is per site, and only with your approval.

06

Keep plain files

A project is a folder of Markdown, .py, .ipynb and CSV that Obsidian, VS Code and Jupyter open too.

Meet Ros, the planner that lives on your laptop.

Ros runs on NVIDIA Nemotron, small enough to download from the Agents page and run on your own machine. It breaks your question into tasks, picks an agent for each, estimates the cost and asks before anything starts.

NVIDIA Nemotron 3 Nano Runs locally No account needed
Does AI adoption help Dutch SMEs grow?
Ros proposes
1Collect sources on SME AI adoptionClaude Code · ~$0.40
2Fit growth against AI useCodex · sandbox
3Note where sources disagreeRos · local, free
Unknown: no panel data before 2021Start all three

Yours, on your own machine.

ros.ai is open source under MIT. It runs on your computer and talks only to the model and search providers you connect.

  • Served on 127.0.0.1. Nothing listens beyond your machine.
  • Keys stay server-side. The browser never sees them.
  • Spending limits per agent. Agents pause when a budget runs out.
  • Local inference. Ros and Ollama models never leave your laptop.

Simple pricing

Open source
Free

The whole app on your machine. Bring your own model keys or run models locally.

Get ros.ai
Organisations
Hosted

ros.ai run for your team, with shared projects, central budgets and admin controls.

Talk to us

Common questions

Which agents can I use?

Ros runs locally. Claude Code and Codex connect with your own keys or logins, Ollama models run on your machine, and any agent that speaks MCP can join through the ros_* tools.

Does my data leave my computer?

Only the requests you send to providers you connect. Projects, keys and the sandbox stay on your machine.

What do I need to run it?

Node 24. ros.ai sets up its own Python on first run. Ollama only if you want other local models.

Is it really free?

Yes. The app is MIT licensed. Organisations pay only if they want us to host it.

Run your agents in the open.

git clone https://github.com/arkdong/ros-ai && cd ros-ai
npm ci && npm run build && npm start

Then open http://127.0.0.1:4310 · needs Node 24

Or try the live demo