Animated WaffleDocsLog in

Account CLI

Use Aniwaffle from your terminal

Install the official CLI, sign in to your Aniwaffle account from your own browser, then list voices, generate speech, and create API keys from a local terminal or a trusted remote agent such as OpenClaw. Google sign-in stays with you; the CLI receives a revocable Aniwaffle credential.

Before you start

Sign in once, then choose a workspace

Install the @aniwaffle/cli npm package and run the aniwaffle command. The account login follows your current Web permissions across workspaces. List and clone voices, generate speech, and manage API keys, members, usage, and billing from the terminal.

  • Runtime: Node.js 20.3 or newer.
  • Permissions: each operation uses the same validation and permission checks as its Web counterpart.
  • Text to Speech: start with voices list, tts generate, and keys create. Available voices, cloning providers, and optional modules follow the selected workspace's configuration. Commands for disabled products or providers are rejected; CLI login does not expand your invitation's permissions.
  • Trusted host: authorize a machine you control. The credential is persistent and revocable; it never grants Google account access.

Setup

Install the CLI and get started

Run this on the coding agent's machine. The global installation provides aniwaffle in your terminal. The npm exec alternative below uses npm's cache instead.

bash
npm install -g @aniwaffle/cli
aniwaffle
aniwaffle login
aniwaffle organizations list
aniwaffle organizations use -f organizationId='<WORKSPACE_ID>'
aniwaffle voices list
aniwaffle tts generate -f voiceId='<VOICE_ID>' -f text='Hello.' -f outputFormat=pcm_24000 --output hello.wav
aniwaffle keys create -f label='Server integration'

tts generate saves audio with your account login. keys create returns a workspace awp_ key for your own server; it is shown only once, so store it privately.

On the first interactive run, aniwaffle asks whether to install its optional Prompt Lab skill for Codex and remembers your choice. Text to Speech does not need it. JSON, help/version, piped commands, and CI never wait for this question.

Running aniwaffle without arguments also shows a short getting-started guide. Lists are tables; account status, login prompts, results, and errors are readable text. Use aniwaffle tts --help to see speech commands, aniwaffle tts generate --help for one command, and aniwaffle --version to check your installation.

Automation: add --json for the previous JSON success/error envelope and login authorization events. This is required for scripts upgrading from CLI 0.4 or earlier; piped output also defaults to readable text. --output file exports retain their existing JSON or audio format.

For a trusted remote coding agent, cache the package first:

bash
env -u WAFFLE_AGENT_TOKEN npm exec --yes --package @aniwaffle/cli -- aniwaffle --help --json

For subsequent commands, --offline uses that cached package. If the cache is missing, repeat the setup step. Keep --json on automated commands.

CLI 0.10.0

Move to the aniwaffle command

bash
npm install -g @aniwaffle/cli
aniwaffle status

Replace waffle with aniwaffle in your scripts. For npm exec, use --package @aniwaffle/cli -- aniwaffle. Existing logins, WAFFLE_* environment variables, and config files under animated-waffle/ are reused; you do not need to sign in again.

The legacy CLI package provides waffle as a compatibility command. It receives matching updates for one month from November 9, 2026; updates stop on December 9, 2026 at 00:00 Asia/Shanghai. Its dependency is pinned to a specific @aniwaffle/cli version, so the final legacy release stays frozen instead of silently tracking later versions. Installed legacy commands keep working; there is no runtime shutdown. New features and fixes after the cutoff ship only in @aniwaffle/cli.

After moving your scripts, remove the legacy CLI package from your global npm installation. The two packages own different command names, and uninstalling the compatibility package does not delete your saved login.

The unscoped aniwaffle npm name is reserved; the maintained package is @aniwaffle/cli. If you installed the unscoped package, run npm uninstall -g aniwaffle before installing @aniwaffle/cli, because both provide the same command. Other Aniwaffle npm packages also moved to the @aniwaffle scope.

Device authorization

Authorize from your own browser

For OpenClaw or another remote chat-driven agent, run the following command in the remote environment. Use the origin from its Dashboard setup; production is shown here.

bash
env -u WAFFLE_AGENT_TOKEN npm exec --yes --offline --package @aniwaffle/cli -- aniwaffle login --json \
  --origin 'https://aniwaffle.com' \
  --no-open --input '{}'

--no-open avoids trying to open a browser on the remote host. --input '{}' prevents the command from waiting for JSON on stdin. On your own desktop, omit --no-open if you want the CLI to open the browser automatically; the approval flow is otherwise identical.

  1. Keep the login process running. With --json, the CLI writes an authorization_required JSON event to stderr containing verificationUriComplete, verificationUri, userCode, and expiresIn. A remote agent should show the user the verification link, matching code, and expiry from this event, while leaving the process running to poll for approval. Do not hide stderr or report login success yet.

  2. The user opens the link on their own device. Sign in to Aniwaffle with Google if prompted, then select Continue, check the account, client (animated-waffle-cli), scope (account:cli), and code, and select Approve. Only approve a request you just initiated on your own trusted agent. Use Deny for an unexpected request.

  3. Wait for the CLI result. The CLI receives the approval through outbound requests to Aniwaffle; no localhost callback, SSH tunnel, or inbound port on the remote host is needed. It exchanges the temporary platform session for a revocable account credential, closes that session, and saves the credential privately. The account credential has no fixed expiry. Every operation checks your current membership and permissions; removing access takes effect without waiting for the credential to expire. If the request expires, run login again and use the new link and code.

  4. Verify the account. After login exits successfully, run status and check the returned user and origin. Select a workspace before running workspace commands. If it reports an existing login, inspect status first; do not revoke or replace another task's login without the user's agreement.

bash
env -u WAFFLE_AGENT_TOKEN npm exec --yes --offline --package @aniwaffle/cli -- aniwaffle status --json

This authorizes the CLI to use Aniwaffle; it does not sign the remote agent into the user's Google account or grant access to Gmail or Drive. Never ask the user to paste a Google password, cookie, access token, or credential file into chat. Only the verification link and matching user code belong in this approval exchange. The remote runner must support a continuing process and access to its output; a chat interface alone is not a CLI integration.

Choose your workspace

bash
npm exec --yes --offline --package @aniwaffle/cli -- aniwaffle organizations list --input '{}'
npm exec --yes --offline --package @aniwaffle/cli -- aniwaffle organizations use --input '{"organizationId":"<WORKSPACE_ID>"}'
npm exec --yes --offline --package @aniwaffle/cli -- aniwaffle voices list --input '{}'

organizations use saves the workspace for later commands. Use --org <WORKSPACE_ID> for a single operation without changing the saved selection. Switching workspaces does not create another login.

After authorization

Commands and Web equivalents

Use --org <WORKSPACE_ID> for one operation, or organizations use to save a workspace. WAFFLE_ORGANIZATION_ID overrides the saved selection. A saved login cannot be sent to a different origin. --input @file.json reads a file, -f name=value adds a typed JSON field, and --output file saves the result privately without overwriting an existing file. Results and errors are readable text by default; --json preserves the machine-readable protocol. Mutations are never automatically retried.

Web operationCLI commands
Account and preferencesstatus, account get, account product, account nsfw
Workspacesorganizations list, use, get, create, update, leave
Members and invitationsmembers list, role, remove; invitations list, mine, create, cancel, accept, reject
CLI loginscredentials list, revoke
Voice catalog and samplesvoices list, create, update, delete, samples, upload, history
Provider voices and cloningvoices search, import, prepare-provider, resolve-model, set-default, set-provider-enabled, add-models, clone, jobs, variant
Recording setssample-sets list, create, upload, update-recording, clone
Speech generationtts tag, tts generate --output audio.wav
API keyskeys list, create, revoke
Usage, credits, and auditusage summary, usage session, billing history, audit list, audit get, performance get
Voice providersproviders status
Platform settingsmodels get, update; admin organizations, account, users, controls, credits, unlimited, nsfw, and audit commands
Platform invitations and alert routingadmin invitations, invite, revoke-invitation, alert-routing, update-alert-routing
Direct authenticated request/downloadapi --input '{"path":"/api/...","method":"GET"}' --output result.json

TTS uses the selected Voice's supported output formats. PCM results receive the same WAV container as Web playback, preserving their sample rate.

For an invited Waffle TTS workspace, choose a ready voice from voices list, then generate a file:

bash
aniwaffle tts generate -f voiceId='<VOICE_ID>' \
  -f text='こんにちは。音声生成の動作を確認しています。' \
  -f outputFormat=pcm_24000 --output hello.wav

This command streams speech using your account login. For a server integration, create a separate workspace key with keys create -f label='Server integration' and follow the HTTP and WebSocket examples. The awp_ key is shown only once; store it privately.

voices prepare-provider accepts {"voiceId":"<VOICE_ID>","provider":"fish_audio"} and returns ready or cloning with the existing or new job id. Concurrent requests share one provider clone. voices resolve-model adds "modelId":"s2.1-pro" and returns the exact route id, reusing the ready provider voice. Repeating it is safe and does not change the API default; voices set-default explicitly changes that default, while voices set-provider-enabled accepts the voice, provider, and enabled boolean. The older voices add-models batch operation remains supported.

billing history and members list support --paginate. Voice uploads accept local file paths; sample-sets upload accepts a files array.

bash
aniwaffle billing history -f source=credits --paginate --output credits.json

Use --json when another program consumes command results. Use --input with one JSON object, or provide that object on stdin.

Mutating commands, such as creating or revoking keys, cloning voices, and changing voice defaults, require the user's authorization. They change remote state and are not setup or verification steps.

Credential lifecycle

Revoke access when it is no longer needed

The saved awc_ credential represents your Aniwaffle account across its workspaces. It is valid until revoked and follows your current permissions. On macOS and Linux, the CLI saves it in animated-waffle/credentials.json beneath WAFFLE_CONFIG_HOME, otherwise XDG_CONFIG_HOME, otherwise the user's .config directory, with file mode 0600. Keep it out of repositories, logs, chat, and shared artifacts. It is not a workspace awp_ key and should not be used in API requests or TTS examples.

To end this login, revoke the credential and clear the local state:

bash
env -u WAFFLE_AGENT_TOKEN npm exec --yes --offline --package @aniwaffle/cli -- aniwaffle logout

If remote revocation fails, the CLI keeps local state so logout can be retried. An account credential the server already rejects as invalid is cleared locally. Follow its error message and retry; deleting a local file alone does not revoke remote access. New interactive setups do not need the legacy WAFFLE_AGENT_TOKEN compatibility override.

To revoke a login on another machine, sign in on a trusted machine and use credentials list to identify its prefix and ID, then credentials revoke --input '{"credentialId":"<CREDENTIAL_UUID>"}'. Only your own CLI credentials are listed or revocable. This never prints their secrets. Revoking the current credential also removes its saved local login.