Skip to main content
Mirage includes twelve command-line programs for working with services, Git repositories and search mounts inside a workspace. Each program’s --help describes the commands and flags implemented by Mirage. The host mirage CLI connects to and manages workspaces. The programs on this page run in those workspaces through ws.shell(...), mirage shell, or another access method.

​
Supported CLIs

All twelve are available in Python and TypeScript on Node. The TypeScript Node package provides hf and himalaya; the other ten live in core. These are Mirage implementations of the listed command trees, not installed copies of the upstream binaries. Each reference links its upstream project or service API and documents its supported subset, limits and relevant licenses. See compatibility and licenses for the distinction between matching behavior and reusing code.

​
Register a CLI

Account CLIs need credentials but do not need a service mount:
registerCli(name, spec, config) validates the config object with the spec’s schema. git and search have no account config; register their GIT and SEARCH specs without the third argument. Two installations of the same spec under different names can use different accounts. Registration is a host operation, not an agent shell command. The equivalent YAML entry is:
YAML config fields use snake_case in both languages. Setup pages cover credentials and imports for every program.

​
Discover functionality

man and --help render the registered spec. Unknown commands and flags are refused; wording and exit codes follow the program’s declared style. Use each reference for command-specific errors rather than assuming one exit code for all CLIs. Visible installed programs have read-only stubs under /usr/bin, so which slack returns /usr/bin/slack. The stub describes the command without exposing its config or credentials. Executing that path uses the same command and policy checks. Changing PATH does not change Mirage’s virtual command lookup; host and remote runtimes use their own executable paths.

​
Use CLIs with the VFS

Credentials may be shared between a CLI and a VFS, but neither registers or locates the other. A CLI write uses service IDs, while file arguments resolve against the session’s working directory and retain PathSpec semantics. After an account CLI write, Mirage expires the caches of every mount the CLI’s client may reach, including answers a mount’s client keeps itself (a vector store’s search results), so later reads can observe the change. A client that names what it reaches narrows this: gh reaches only GitHub mounts on its API base. Mount modes and path policy govern workspace file operations. A read-only Slack mount does not disable slack send-message; use command policy to restrict service actions. The service’s own credentials and permissions still apply to its API calls.

​
Add a custom CLI

Use the authoring guide to define a typed command tree, register multiple accounts, work with VFS paths, or run a script CLI. It also covers config validation, snapshots and the shared command grammar.