---
title: "Computer & previews"
description: "Use your session’s files, configuration, environment templates and app previews."
---

Each session runs on an isolated Beeuni Computer. It holds the agent’s running processes and working files. The session also keeps a saved conversation, so it can continue after a stop or recover onto a new Computer. See [Sessions](/docs/work/sessions) for lifecycle and retention rules.

<a id="files"></a>

## Files and branches

For an agent with repository access, `/workspace` contains the project checkout on the session branch. Files uploaded before the first prompt appear under `/workspace/incoming`. Commit and push changes you need to keep. Stopping preserves the disk; deleting or losing the Computer does not.

Use the session’s file browser, `beeuni sessions files` or `beeuni sessions cp` for live files. Use the project’s Files page for committed files and `beeuni sessions attachments` for saved attachments. These stores are distinct; saved attachments and repository files do not require the Computer to run.

<a id="config-convergence"></a>

## Apply configuration updates

<a id="where-a-box-gets-the-release"></a>

The session runs agents, skills and tool permissions from its base branch. An edit only in `/workspace` does not change the active configuration. Push the change to the base branch, then choose **Reload config** or use the CLI. Automatic configuration updates wait until the active turn ends; an explicit reload during a turn can require stopping it.

```bash
beeuni sessions reload <session_id>
beeuni sessions reload <session_id> --no-repo
```

Reload reports its progress and validates the new configuration. If it fails, the header shows **Config failed to load**. Correct the project configuration and retry. Beeuni can keep the last working configuration while the problem is resolved.

If a configuration removes or renames the session’s agent, Beeuni can switch to the project’s default agent only when access allows it. Otherwise, restore the agent or ask a project manager to update its access. A reload does not merge base-branch code into the session branch; use **Ask Agent: Sync Branch & Reload** for that.

<a id="the-image-and-templates"></a>

## Environment templates

Use **Customize → Settings → Sandbox templates** or `sandbox.templates` in `beeuni.yaml` to choose the environment and resource size. New sessions use `sandbox.default` when set, otherwise the platform default. See the [manifest reference](/docs/project/manifest#sandbox) for supported fields.

An image or resource change applies to new Computers, not the one already running. Store credentials as [project secrets](/docs/project/secrets), not in an image or repository. Only secrets granted to the session are delivered according to their exposure setting.

<a id="docker-in-the-sandbox"></a>

## Docker in the sandbox

For tasks that run Docker, set `container_runtime: true` on the template and install Docker in its image. This setting applies to new Computers. Use `cpu`, `memory` and `disk` on the template to set their resource size.

<a id="ports-and-previews"></a>

## Ports and previews

Start an app in the Computer, then open its port from the session’s preview panel. The CLI can print a preview URL or forward a port to your machine. A preview request wakes a stopped Computer. Do not share a URL containing your sign-in token; create a public share link from the session instead.

```bash
beeuni sessions preview <session_id> 3000
beeuni sessions forward <session_id> --port 3000
```

A preview shows the app at the selected port. If it is empty, check that the server is running and listening on that port. Start long-running servers in `tmux` so they survive the agent’s tool call. Public share links cannot expose Beeuni’s reserved service ports.

<a id="troubleshooting"></a>

## Troubleshooting

If agents or skills are out of date, check the base branch and reload configuration. If the Computer was lost, use **Continue session**. If a turn ends because of memory pressure, reduce concurrent heavy processes or increase the template’s memory for new sessions. Use the saved session log to inspect what the agent completed before retrying.