Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
13 changes: 13 additions & 0 deletions docs/inspector/computer.mdx
Original file line number Diff line number Diff line change
Expand Up @@ -116,6 +116,19 @@ Manage them from **Change** on the image strip.
any one computer's disk — see [Data & persistence](#data-and-persistence).
</Note>

## Cloud sandboxes

Several surfaces run computer commands in **MCPJam cloud sandboxes** — disposable machines that are provisioned on demand and torn down when the session ends. These sandboxes are always cloud-hosted, regardless of where the inspector itself runs.

| Surface | Sandbox scope |
|---|---|
| **Chatboxes** (environment with pinned image) | Per conversation |
| **Swarm sessions** | Per attempt |
| **Eval iterations** | Per iteration |
| **User Testing sessions** | Per conversation |

Sandbox images (defined under **Computer → Images**) apply to all of these cloud runs. They never apply to the machine running the inspector itself.

## Chatbox sandbox

When a chatbox is backed by an environment that has a pinned computer image,
Expand Down
4 changes: 4 additions & 0 deletions docs/inspector/evals.mdx
Original file line number Diff line number Diff line change
Expand Up @@ -194,6 +194,10 @@ The first run of a suite saves the set of MCP servers used as a frozen snapshot;

In suite settings, the **Computer environment** select lets you pin a built Docker environment so every eval iteration boots a fresh sandbox from the same image. This keeps results comparable across runs and edits — each iteration gets an identical starting state.

<Note>
Eval computer commands always run in **MCPJam cloud sandboxes** — never on the machine running this inspector. If this inspector cannot reach MCPJam cloud sandboxes, **Run all** is disabled for suites that pin a sandbox image.
</Note>

- Choose an environment from the list. Environments that have not finished building are labelled **(not built)**; selecting one will cause the run to fail immediately.
- Choose **None (default image)** to run without a pinned environment.
- The agent receives a `bash` tool inside the sandbox.
Expand Down
Loading