# Your computer

A place for your agents to work, keep context, and improve through changes you review.

Canonical page: https://beeuni.ai/docs/cloud-computer

Beeuni is a computer for your agents. They use your tools, produce real
work, and build on the skills and context you keep.

The computer is more than a place to store files. It brings together the
agent's instructions, skills, tool access, and work history. The model does
the reasoning; your project holds the context it works from.

## What carries forward

Your project keeps the agents, skills, instructions, and memory you have
configured, along with the code and materials you merge into it. Sessions
record conversations and progress so you can return to the work.

Each session runs on its own isolated computer and branch. Work in one
session does not automatically become another session's context. To carry
file changes forward, review and merge the agent's change request. Connected
databases and tools keep their own data and permissions.

See [Projects](/docs/project), [Sessions](/docs/work/sessions), and
[Change requests](/docs/work/change-requests).

## Learning from work

After a task, an agent can propose a better skill, a clearer instruction, or
a useful addition to memory. You inspect the change and decide whether to
keep it. Future sessions can use the updated project.

That is the learning loop behind recursive self-improvement (RSI) in Beeuni:
work, propose a change, review it, and use what helped in the next task.
Improvement means changing how the agent works, not retraining the model's
weights or guaranteeing a better result after every run.

## Change the model, keep the project

The agent's skills, instructions, and project files are separate from the
model. You can choose a different configured model without rebuilding that
context. Models and harnesses can handle conversation history differently;
switching them does not promise an identical continuation of a live session.

See [Models](/docs/project/models) and [Harnesses](/docs/work/harnesses).

## Stay in control

Agents work within the access you give them. Tool rules can allow a call,
ask for your approval, or block it. Agents can work in parallel, each in its
own session, and propose changes to the shared project.

Pausing a session and keeping project state are different things. Sandbox
lifecycle settings determine how long its running environment remains
available. Committed work and reviewed project changes provide the lasting
context; they do not depend on one machine staying alive forever.

See [Sessions](/docs/work/sessions) and [Connectors](/docs/connect).

## Start

```bash
curl -fsSL https://beeuni.ai/install | bash
beeuni init
beeuni ship
```