Your own AI agent — ready in minutes

A real Machine — not a chat window

Your agent gets its own private Linux computer environment in the cloud: a web UI, a terminal, a file system and a full web browser. It can develop, test, run and push code and applications — not just talk about it. Write a script, run it, see the result, fix the bug and run it again. Download your finished applications easily — or use cloud providers like Google to store your files! Did I mention it? Of course you can connect your email, social media and more.

Agent Zero is capable of being the support-bot for your business, and at the same time easy enough to help you learn how to use a computer!

Dark mode or light mode — your choice

The interface follows you, not the other way around. Switch between dark and light themes whenever you like — your chats, files and settings stay exactly the same. Drag the slider below to see the same view in both themes.

Dark Light

Works with real documents — and writes with you

The agent works alongside you on documents, texts, spreadsheets and presentations — instead of looking to a webpage for help on how to do it. Ask it to draft a plan in the open document and watch the text appear in the shared context, while you keep working on another section. You and the agent learn to work together, on the same documents. Make plans, TODOs, meeting notes, handoffs — anything you can do on a computer, the agent can do! It has its own dedicated search engine, aka SearXNG, which it uses to find information, query documents and more. And it can also use its Chromium (Playwright) browser: so you can see how your webpage is looking, fill in complex forms online — whatever you do, you are covered! Want to see the Python application you are working on with the agent? Activate the plugin and connect to the desktop with WebSocket. (FYI: keep in mind the heavier you use memory and processes, the more resources you will need)

Works with your own computer too

Your agent is not locked inside a browser window. When you allow it, it can reach tools on your own machine — so a task that starts in the cloud can finish on your desktop. You decide what it may touch, and you can withdraw the permission whenever you want.

Be your own software developer

When you start a new project, you can choose a public GitHub repository — or connect to your private repo with your own git login — and it is cloned directly into your project folder, ready to work on. Find an open-source project you want to work on, and make your own custom version of any software, tailored to your needs!

Builds its own tools and skills

When your agent is missing a tool, it can build one itself — or browse hundreds of community plugins and add new amazing abilities, like "NeuralMind" for persistent codebase memory, or "Parallel Swarm" to run multiple agents in parallel with fan-out/fan-in execution that lets agents share findings mid-execution. "AutoResearch" is an autonomous code-optimization experiment loop: it generates hypotheses, edits code, runs benchmarks, evaluates results and iterates until convergence! Then there is Dreaming, document handling, integrations and much more — and these are just some of the tools available. You can easily bring tools to any agent by telling Agent Zero to make it compatible. Plugins get a safety check that runs a full security scan of both new and old plugins! You get an assistant that grows with your needs, without opening the door to strangers.

Multimodal agent

Agent Zero uses multiple models to be both more cost-efficient and faster. You choose one main reasoning model — the "big brain" that organizes everything — and a smaller, faster utility model that handles tool calls, memory consolidation, web search and more. That saves a lot of API tokens! And if your reasoning and utility models have no vision — and you do not want the value-versus-vision tradeoff — you can preset a separate vision model that steps in whenever a task needs to see. This multimodal framework is what makes Agent Zero truly unique — and revolutionary!

It gets smarter the more you use it

Your agent remembers what matters. As you work together, the long and unimportant gets compressed over time — what is important stays. The more you use it, the more robust and capable it becomes at exactly what you do. And feel free to play and experiment: if you are curious, just try things out — that is how you discover what it can really do.

Branch a conversation from any point

Whenever you want, you can split the conversation: pick any earlier point in the chat and press a button under the message — a new branch of the conversation grows from there. Explore a different idea of what AI can be — and keep your memories, history and money for yourself.

Time travel for everything it does

Work your agent does is recoverable by design. See exactly what changed, compare versions, inspect an earlier state and roll back whenever you want. A task that went wrong is never lost work — it is a few clicks back in time. That makes it safe to experiment: let the agent try something, knowing you can always return to what worked.

Choose from hundreds of LLMs — use it with pseudonymity

Hundreds of models are at your fingertips — from light workhorses to heavy frontier models at $50 a million tokens! You pick the ones you want. We will not lock you to any model, but we will give strong recommendations and presets that make it easy to use cheap but strong models! Your agent is also multi-model by design: a stronger model handles reasoning, planning and orchestration by default, while smaller, more efficient models take care of memory consolidation, tool calls, writing commands and searching the internet. Splitting the work between a reasoning model and a utility model saves a good amount of tokens — without giving up any capability (you actually gain quality and speed).

Your privacy stays part of our focus: we save only what is needed — your name and contact details do not follow along, or get passed around! We are the only ones communicating with the providers (we look for good API-token deals for our users). We monitor which providers give the best value versus stability every month! And since we do not share who our users are with the providers, it is harder to trace you as a person — and you are never locked to a single provider.

Multiple agents working together

Big tasks get split up: your agent can delegate tasks — research, coding and analysis — to subordinate agents that work in parallel (deploy as many as you like) and gather the results when they are done. You steer the flow, optimize the workflow and make your own custom prompts and feedback loops, so the agent is as smart as possible for as little as possible! (Or you just use a plugin from someone else who has more time than you.) Either way, it is free — it is open source :)

Open by design. Shielded for you!

Your agent runs in its own shielded environment with access to only what it needs. It is separated on its own encrypted network, and you decide who gets access to the agent. We have implemented 2FA, so you can be sure nobody breaches your orchestrator agent — the one that controls your YouTube empire or the support bots for your business. You should feel safe! That is my mission — and to do it for cheap.

If your agent needs extra resources, the resource limits are not set in stone! You have higher burst available, and the server has automatic load balancing which gives you more resources when they are free. If you need even more security, or dedicated power — say for heavy 3D work in Blender or projects like that — the Professional plan is for you: get a custom domain, connection through a personal virtual network (VPN) which we provide, or a 100% dedicated server for security reasons — or if you want a team of agents on one server. Custom servers are provisioned within 24–48 hours.

*Regular servers from Small to Large on nxm.nu are up within minutes! (As long as we are not sold out!)