Why Unbound runs bots on your own computer
Last reviewed
Unbound gives each bot its own working environment on hardware you control, while you keep using your desktop as normal. Run the bots and a local model on your PC, or on a server you operate. The model processes your prompts and the bot’s context on that machine, with no hosted AI provider involved.
Why keep your AI team off someone else’s cloud?
An AI team becomes useful by connecting to things: email, calendar, files, code, browsers, and the dozens or hundreds of connectors and plugins people add over time. With a hosted agent, every one of those connections runs through the provider’s cloud. Each connector’s credentials, the data it returns, and the actions bots take with it pass through infrastructure you do not operate, under that provider’s retention, logging, and access policies.
- One less party in the path. Connector requests go from your machine straight to the service you connected, not through an agent host’s servers.
- Your credentials stay with you. Tokens and API keys are stored encrypted on your machine, not pooled with other customers’ in a provider’s cloud.
- Less exposure as you add connectors. Every new plugin on a hosted agent widens what that provider can reach. Locally, it widens only what your own machine can reach.
- Data a host never holds cannot leak from it. A breach, legal demand, or policy change at an agent provider does not reach a workspace that provider never stored.
- Your history stays yours. Conversations, memory, and files remain in your workspace when you change models or providers.
Connected services still see their own requests, and choosing a remote model sends that task’s context to the model’s provider. Running locally takes the agent host out of the path; adding a local model takes the AI provider out too. The FAQ covers each boundary.
Local, self-hosted, and cloud: what changes?
Local means Unbound runs on the machine in front of you. Self-hosted means you run it yourself somewhere else, such as a second PC or a server you operate. A managed cloud computer is hosted and operated by the agent provider.
A bot’s “computer” is a working environment, not a separate physical machine for every bot. Some setups use Docker to keep those environments apart. In every case you control the host, the stored workspace, and the connected accounts. See the Security & Data page.
Choose local models for private AI processing
Use Ollama, LM Studio, or vLLM to run a model on your own hardware. When the bot, the model, and the tools a task uses all run on your device without reaching out, that task’s prompts, files, and context stay entirely on it.
You can also bring your own API keys for any built-in or custom provider. Choosing a remote model sends it the context it needs for the task. Email, calendar, drive, and websites still exchange data with their own services even when the model is local. The FAQ explains those boundaries.
What stays on your machine?
Conversations, bot memory, routines, files, tool output, screenshots, and browser profiles are stored in your workspace or in persistent Docker volumes. Browser profiles can keep you signed in to sites between tasks. The workspace, accounts, and backups are yours to manage, like any other local data.
For example, a bot using a local model can summarize documents on your disk and draft a file with local tools without that document content ever going to a cloud model.
What needs to stay running?
The machine running Unbound has to be on and reachable for work to continue. A self-hosted server can be a different machine from the laptop you control it from, so the laptop can sleep while the server keeps working. If the host itself sleeps or is powered off, its tasks pause until it is back.
How are credentials protected?
API keys and other secrets live in a credential vault encrypted with AES-256-GCM, so copying the files does not expose them. On the desktop, the vault key is protected by your operating system’s secure storage by default. On a server with no desktop, you supply and protect the key yourself. If you reach the server remotely, follow the secure deployment guidance for authenticated access and encrypted transport.
Where should I start?
Download Unbound for Windows (Beta). Running it yourself is free; your AI provider and hosting bill you separately. Follow the setup instructions included with the release.
Read the comparison with Muse, Dots, and Grok Bot if you are deciding between running your own environment and using a managed agent computer.
Frequently asked questions
Why run an AI team locally instead of with a hosted agent?
A hosted agent routes every connector and plugin you add through the provider’s cloud, credentials, retrieved data, and actions included. Running Unbound on your own computer or server takes that host out of the path: connector requests go straight from your machine to each service, and credentials stay encrypted on your machine. Connected services still see their own requests, and a remote model, if you choose one, receives the context it needs.
Can Unbound be self-hosted?
Yes. Run Unbound on your own computer or a server you control. You manage the host, provider keys, connected accounts, permissions, and backups.