LS AI Provider for ElevenLabs

1.0.2

• Rename the plugin to “LS AI Provider for ElevenLabs”, with the permalink ls-ai-provider-for-elevenlabs. The WordPress.org plugin directory requires a plugin name to begin with a distinctive identifier, and “AI Provider” on its own is a generic description; “LS” is the author’s initials
• Rename the bootstrap to ls-ai-provider-for-elevenlabs.php and the text domain to ls-ai-provider-for-elevenlabs, both of which have to match the permalink
• Keep the filter, option and transient names (ai_provider_for_elevenlabs_*) and the Composer package name (saarnilauri/ai-provider-for-elevenlabs) unchanged, so existing integrations and Composer installs are unaffected

1.0.1

• Verify the API key against ElevenLabs instead of only checking that one was entered, so Settings > Connectors reports whether the key actually works. A key rejected by ElevenLabs now shows as not connected, where any non-empty value previously showed as connected
• Treat a key scoped without the Models permission as working rather than broken. Such a key still drives text-to-speech, and the plugin already falls back to its built-in model list, so the connection check accepts it. The verification result is cached for 15 minutes, under a name derived from a hash of the key rather than the key itself
• Stop reading the core Connectors option (connectors_ai_elevenlabs_api_key) directly. WordPress passes the key to the AI Client itself, and the plugin already re-wraps it for the xi-api-key header ElevenLabs requires, so the direct read was redundant. This resolves a WordPress.org review finding about plugins handling AI Client credentials

1.0.0

• First release from the WordPress.org plugin directory
• Rename the plugin bootstrap from plugin.php to ai-provider-for-elevenlabs.php, so the plugin headers live in the file matching the plugin slug as the WordPress.org directory expects. Sites that installed an earlier build from the GitHub ZIP have to reactivate the plugin once after updating
• Document every ElevenLabs endpoint the plugin can call, what is sent to it and when, in a new “External services” section of the readme
• Correct the readme’s claim that a separate “PHP AI Client plugin” has to be installed: the AI Client is an SDK, bundled in WordPress 7.0 and later
• Redraw the provider logo shown on the Connectors screen from the official ElevenLabs symbol, at the proportions and clear space their brand guidelines specify
• Add the WordPress.org directory banner and icon, built from the official ElevenLabs artwork in their monochrome palette, a WordPress Playground blueprint for the directory preview, and a Plugin Check run in continuous integration

0.4.0

• Long-form narration: text beyond the model’s per-request character limit is split on paragraph and sentence boundaries, narrated across several requests carrying neighbouring text for prosody, and returned as one audio file (contributed by Jake Spurlock)
• Automatic voice selection: when no outputSpeechVoice and no explicit default are configured, a voice is discovered from the account’s own voices, preferring premade ones, before falling back to “George” (contributed by Jake Spurlock)
• Voice directory: migrate from the deprecated /v1/voices endpoint to /v2/voices with full pagination, and cache the voice list per API key for 15 minutes (contributed by Jake Spurlock)
• Honour all customOptions in both the text-to-speech and sound generation models (e.g. language_code, seed, speed, apply_text_normalization); previously most keys were silently dropped (contributed by Jake Spurlock)
• Read each model’s real per-request character limit from the live /models response, seeded from measured values, and add eleven_v3 to the fallback model list (contributed by Jake Spurlock)
• Fix the voice directory staying unusable for the rest of the request when built before credentials were available (contributed by Jake Spurlock)
• Show the provider in the connector UI as “ElevenLabs” with a description and the official logo (contributed by Jake Spurlock)
• Exclude local credentials (.env, .wp-env.override.json) and dev artifacts from the release ZIP, verified by a CI canary check (contributed by Jake Spurlock)
• Add continuous integration (unit tests on PHP 7.4-8.4, phpcs, PHPStan, packaging leak check), a local wp-env environment, the GPL-2.0 license text, and fix test-suite autoloading on case-sensitive filesystems

0.3.0

• Declare inline outputFileType support for text-to-speech and sound generation models, so support checks like isSupportedForTextToSpeechConversion() pass when callers request inline output (fixes compatibility with the WordPress AI plugin’s Text to Speech experiment)
• Fall back to the premade “George” voice when no outputSpeechVoice is configured, instead of throwing an exception; the default can be overridden via the ELEVENLABS_DEFAULT_VOICE_ID environment variable or constant, the ai_provider_for_elevenlabs_default_voice_id option, or the ai_provider_for_elevenlabs_default_voice_id filter

0.2.0

• WordPress 7.0 compatibility: read the API key from the core Connectors option (connectors_ai_elevenlabs_api_key, Settings > Connectors)
• WordPress 7.0 compatibility: move the xi-api-key authentication restore hook to init priority 21, after core’s Connectors credential pass (init 20) which overwrites provider auth with generic Bearer authentication
• Fix API key validation on the WordPress 7.0 Connectors page: the availability check now also considers the registry request authentication and the Connectors/legacy credential options, instead of only the ELEVENLABS_API_KEY environment variable or constant