Multi-AI Chat v0.3.0

Experimental Meta AI standby

• Adds Meta AI as an optional provider that stays off by default. The original four remain active. In Settings → Standby provider, one of them can be replaced by Meta AI, and exactly four providers stay active.
• Requests access to https://www.meta.ai/* and https://meta.ai/* only when the user activates Meta. Declining leaves Meta on standby. meta.ai is not a required host permission, so existing users are not forced to re-approve the extension on update.
• Registers content/meta.js dynamically after that access is granted. There is no static Meta content script.
• Treats a usable composer as ready even when leftover login controls are still on the page. An inert composer or an explicit login wall still means sign-in.
• Rejects a composer or control that is gated by data-disabled="true" or drawn at opacity: 0, so a form that is disabled or mid fade-out is no longer read as ready.

Workflow, Settings, and input

• Keeps a late result, provider URL, or send failure from the previous request out of a new chat. Conversation storage writes the latest snapshot, retries a failed read instead of wiping history, and skips malformed rows.
• Names providers that are not ready before a workflow or a partial Free send, opens the selected unready providers from the readiness hint, and can select only the ready active providers. Free mode stays selected when the panel closes.
• Keeps IME confirmation from sending a draft. Exposes mode selection, shows keyboard focus, and keeps Settings focus inside the dialog until Escape or close.
• Keeps Send and the other input actions visible in a short panel, and gives the prompt a stable localized label.
• Ignores a Settings token load or save that finishes after that dialog has closed. A failed load or save stays on screen and can be tried again.

Five interface languages

• Covers standby, Meta permission denial, readiness, Settings, and input copy in English, Traditional Chinese, Japanese, German, and Korean.

Known limits

• Meta AI is experimental and off by default. Authenticated Meta send and receive has not been exercised on a real signed-in page in this extension, and the permission prompt, the dynamic content-script registration, and browser-restart persistence have unit-test coverage rather than a browser run.
• If Meta AI misbehaves for you, please open an issue — that feedback is how this provider will graduate from experimental. The other four providers are unaffected by anything in this section.

Chrome Web Store package

Upload multi-ai-chat-store-v0.3.0.zip as the update for the existing Chrome Web Store item. SHA-256: d7e976872d2e19cf8867ea7562c263d7de31f175e706ea44b67a204db3233d80.

Validation: 231 tests, TypeScript, production webpack build, and version consistency all passed, with CI and CodeQL green on every change in this release.

Multi-AI Chat v0.2.3

Grok completion and Retry reliability

• Supports the current Grok textarea composer as well as the legacy ProseMirror cohort.
• Recognizes the exact current user turn and only accepts assistant output that follows it.
• Ignores hidden or historical Stop and Thinking elements, while keeping the provider connected during active generation.
• Scopes recovery, timers, response tracking, and Stop to the exact request so Retry cannot be poisoned by the failed turn.
• Serializes only the answer body instead of Grok thinking UI.

Chrome Web Store package

Upload multi-ai-chat-store-v0.2.3.zip as the update for the existing Chrome Web Store item. SHA-256: 8fea5b0edbed1b7d818898febef7e2bf4151a0e180ca9dcbc452de1759bd9cca.

Validation: 100 tests, TypeScript, production webpack build, version consistency, CI, and CodeQL all passed.

Multi-AI Chat v0.2.2

ChatGPT web stability

• Keeps one stable owner tab per provider so duplicate ChatGPT tabs and SPA route changes no longer steal or flicker the connection.
• Re-detects remounted ChatGPT composers, stabilizes login status, supports current send/stop controls, and prevents focus-induced page jumps.
• Confirms sends before retrying to avoid duplicate prompts and recovers cleanly from frozen or discarded provider tabs.

Chrome Web Store package

Upload multi-ai-chat-store-v0.2.2.zip as the package for the new Chrome Web Store item. Its SHA-256 is 425bd80abc4618908ef184e75d7ee57e384364e7ed784924deb624a700361919.

Validation: 77 tests, TypeScript, production build, version consistency, dependency audit, CI, and CodeQL all passed.

Multi-AI Chat v0.2.1

Highlights

• Roundtable workflows can now Retry, Skip this turn, or Cancel when a provider fails. Retries use fresh request IDs; skips keep raw provider errors out of later prompts.
• Recovery stays correctly scoped across Side Panel reloads and multiple windows, with stale and late actions rejected.
• Provider URL handling now validates exact HTTPS hostnames, closing the CodeQL findings addressed in #25.
• CI, Dependabot, CodeQL, secret scanning, and push protection guardrails are in place; build dependencies are current and npm audit reports zero vulnerabilities.
• Store documentation now includes explicit Limited Use/privacy disclosures plus reproducible 440×280 promo and full-bleed 1280×800 screenshot assets.

Validation

• Node 22.18.0
• 42/42 tests passed
• Typecheck and production build passed
• CodeQL: 0 open alerts
• Manifest V3 permissions and host permissions are unchanged

Package

multi-ai-chat-store-v0.2.1.zip has manifest.json at its root and is ready for Chrome's Load unpacked flow after extraction or for upload through the Chrome Web Store Developer Dashboard. Verify it with the attached SHA-256 file.

Thank you to @DaveTseng2019 for the display-mode, reliability, accessibility, icon, and documentation contributions included in this first GitHub release, and to everyone who reported issues and tested the extension. 🙏