Skip to main content
anona-compute is one static binary for Linux, macOS and Windows (amd64 and arm64). It is called anona-compute because anona is the Anona Memory CLI; the two have separate sign-ins and settings.

Install

install_cli.sh
The npm package is a small launcher that runs the native binary for your platform. Without Node.js, use either of these instead:
  • Download the binary from the releases page (releases tagged cli-v<version>). Each platform has an archive holding the single anona-compute binary: .tar.gz for Linux and macOS, .zip for Windows. SHA256SUMS lists their checksums. Unpack it and put anona-compute on your PATH:
  • Build it with Go 1.26 or newer. The source is in the public anona-compute-sdk repository, under cli/:
    The binary goes to $(go env GOPATH)/bin; make sure that is on your PATH.

Sign in

Or use an API key, which wins over a browser sign-in:
cli_key.sh

A tour

cli_tour.sh
sandbox (alias sbx) groups the lifecycle commands: anona-compute sbx ls is anona-compute list.

Output and exit codes

--json (or -o json) prints JSON on stdout; messages go to stderr. A list command (list, ps, template list, config list) prints one JSON array, [] when there is nothing to list, so jq -r '.[].sandbox_id' reads it at any length; every other command prints one JSON object per line. Every key is snake_case (sandbox_id, template_id, exit_code), as in the MCP server’s tools. An error is one object on stderr: {"error":{"message":…,"exit_code":…}}, plus status, error_code and request_id when known, and "outcome_unknown": true when a request that creates something failed in a way that does not say whether it was carried out (see Retries). exec and run exit with the remote command’s own exit code, unchanged, so the CLI composes in shell scripts and CI; when the CLI itself fails they exit 125 (as docker run does), and the --json error keeps the detailed code in cli_code. connect exits with the shell’s exit code. anona-compute doctor checks the configuration, the credential and the connection when something does not work.

Retries

The CLI retries a request at most three times and for at most about a minute, waiting Retry-After seconds when the response gives them and backing off otherwise.
  • A Should-Retry header decides on its own: true retries, false never does.
  • Without it, a 429 is retried, since it is refused before anything happens.
  • Without it, a 502, 503 or 504 is retried for reads, deletes and requests that can safely arrive twice (pause, resume, timeout, revoking links).
  • Without it, a 502, 503 or 504 is not retried for a request that creates something: create, fork, a template build, a preview link or a vault secret. Such a response comes from something in front of the API, which cannot know whether the sandbox was already started, and a second create could start a second, billed sandbox. The command fails saying so; list your sandboxes before trying again.
The API’s own errors on create always carry Should-Retry, so a create still waits out a 503 while the platform starts or has no capacity yet. create and resume wait longer than that while the platform has no capacity yet: a 503 with Should-Retry: true and error_code sandbox_capacity_unavailable (a host is starting), sandbox_engine_starting (the platform is waking) or sandbox_size_preparing (a template is being prepared at a new size) is retried for up to --capacity-wait, 10 minutes by default (ANONA_COMPUTE_CAPACITY_WAIT; 0 keeps only the short budget above). Each wait follows Retry-After, held between 1 and 30 seconds with a little jitter, and prints a line on stderr, so --json output stays clean:
resume --capacity-wait, the 1 to 30 second bounds and sandbox_size_preparing need CLI 0.4.0 or later; 0.3.x waits on create only.