Skip to content

Open-source AI browser agent for Chrome and Firefox (monorepo) 🧠

Refresh Chrome and Firefox store listing drafts

Emre Sokulluea7db65fd027
diff --git a/assets/store-listing-chrome.txt b/assets/store-listing-chrome.txt
index d6bfd1f92..fa3f12c84 100644
--- a/assets/store-listing-chrome.txt
+++ b/assets/store-listing-chrome.txt
@@ -1,59 +1,52 @@
----
-CHROME WEB STORE LISTING - WebBrain v0.9.0
----
+CHROME WEB STORE LISTING - proposed English copy, reviewed 2026-10-01
-TITLE: WebBrain β€” Open-Source AI Browser Agent
+TITLE:
+WebBrain β€” AI Browser Agent
SUMMARY (132 chars max):
-Open-source AI browser agent. Use any LLM (OpenAI, Gemini, Anthropic, local models) to read, analyze, and automate web pages.
-
-DESCRIPTION (up to 16,000 chars):
-WebBrain is an open-source browser extension that turns any LLM into a powerful browser agent β€” right from a side panel.
-
-Unlike proprietary alternatives, WebBrain is provider-agnostic: use OpenAI, Google Gemini, Anthropic Claude, or local models via OpenRouter. Your choice, your API key, your data.
-
-TWO MODES:
-
-πŸ” Ask Mode (Default)
-Read-only page analysis. Ask questions about any web page without modifying anything. Perfect for:
-β€’ Summarizing articles and documents
-β€’ Extracting data from tables and lists
-β€’ Analyzing page structure and content
-β€’ Getting explanations of complex content
-
-πŸ€– Act Mode
-Full browser automation with your permission. Let the AI agent interact with pages on your behalf:
-β€’ Click buttons and links
-β€’ Fill out forms
-β€’ Navigate between pages
-β€’ Scroll and read dynamic content
-β€’ Execute multi-step workflows
-
-SUPPORTED PROVIDERS:
-β€’ OpenAI (GPT-4o, GPT-4.1, etc.)
-β€’ Google Gemini (Gemini 2.5 Pro, Flash, etc.)
-β€’ Anthropic Claude (Claude Sonnet, Opus, Haiku)
-β€’ OpenRouter (access to 200+ models including local/self-hosted)
-β€’ Any OpenAI-compatible API endpoint
-
-KEY FEATURES:
-β€’ Provider-agnostic β€” switch between AI providers anytime
-β€’ Side panel UI β€” always accessible, never in your way
-β€’ Ask & Act modes β€” read-only analysis or full automation
-β€’ Configurable agent steps β€” control how many actions the agent takes
-β€’ Per-tab conversations β€” separate chat history for each tab
-β€’ Page inspection indicator β€” see when the agent is analyzing a page
-β€’ Dark mode support
-β€’ Open source (GPL-3.0-or-later for version 33.0.0 and later)
-
-PRIVACY & SECURITY:
-WebBrain runs entirely in your browser. Your API keys are stored locally and never sent to any third-party server. All communication goes directly between your browser and your chosen AI provider. The full source code is available on GitHub for audit.
+Open-source AI browser agent. Chat with pages and PDFs, extract data, and automate tasks with your choice of AI.
+
+DESCRIPTION:
+WebBrain is an open-source AI browser agent for Chrome. Ask questions about pages and PDFs, collect information, and automate web tasks from a side panel beside your tabs. Work in the browser you already use, with your existing website sessions and your choice of AI model.
+
+READ, RESEARCH AND GET THINGS DONE
+
+β€’ Chat with pages and PDFs: summarize articles, understand documents, translate passages, and ask follow-up questions without copying everything into another app.
+β€’ Automate browser tasks: describe what you need in plain language. The agent can click, type, fill forms, navigate between pages, and work through multiple steps.
+β€’ Extract useful data: collect tables, lists, links and page details for research, comparisons and reports.
+β€’ Schedule tasks and watch pages: run a task later or at a recurring interval, and monitor a page for a condition you specify. Your browser must be running for scheduled work.
+β€’ Reuse your workflows: save successful runs, teach a workflow by demonstrating actions, and import or export workflows to use again.
+
+PRACTICAL TOOLS IN THE SAME SIDE PANEL
+
+β€’ Full-page screenshots: capture the visible tab or a full scrollable page and save the image.
+β€’ Bulk file downloads: download multiple linked files in one task, including documents and images.
+β€’ Article and video summaries: read New York Times articles and summarize YouTube transcripts through the included reading tools. Availability depends on the source and connected service.
+β€’ CAPTCHA assistance: connect a supported solving service to help continue tasks blocked by supported challenges. Optional setup and a separate service account are required; charges may apply.
+
+CHOOSE WHERE YOUR AI RUNS
+
+Start with WebBrain's built-in model without an API key. For local inference, connect to a model running on your computer through LM Studio or Ollama. For cloud models, connect your own API key for OpenAI, Anthropic Claude, Gemini or OpenRouter. Additional connections are available in Settings.
+
+STAY IN CONTROL
+
+Use Ask mode for read-only questions and Act mode for browser actions. Review a plan before execution with Plan before Act, manage site permissions, and stop a run at any time. Dev mode adds page inspection and debugging tools.
+
+Your selected AI service processes the context needed for your request. Local models can keep model inference on your computer; connected reading and CAPTCHA tools use online services. The built-in model's Help Improve WebBrain setting is on by default and can be turned off. See the privacy policy for data-use details.
+
+OFFLINE READING, TOO
+
+Optional Apocalypse Mode adds offline AI, a Wikipedia reader and a reference library. Download the models and archives first; on-device AI requires compatible WebGPU support. Download sizes vary by model and archive.
+
+The extension is free to install. AI and connected services may have separate usage limits or charges. Source code is available under GPL-3.0-or-later.
+
+Open the side panel and try: "Summarize this page and list the key facts."
Website: https://webbrain.one
-GitHub: https://github.com/webbrain-one/webbrain
-License: GPL-3.0-or-later (version 33.0.0 and later; earlier releases remain MIT-licensed)
-Author: Emre Sokullu
+Privacy: https://webbrain.one/privacy
+Source and support: https://github.com/webbrain-one/webbrain
-CATEGORY: Productivity
+END DESCRIPTION
-LANGUAGE: English
+EDITORIAL NOTE - not part of the store description:
+This is proposed copy, not a published store update. The title and summary are separate metadata; Chrome also requires manifest changes in a future package. See store-listing-review.txt for evidence and publication notes.
diff --git a/assets/store-listing-firefox.txt b/assets/store-listing-firefox.txt
index 031cb09f3..a17711bf3 100644
--- a/assets/store-listing-firefox.txt
+++ b/assets/store-listing-firefox.txt
@@ -1,72 +1,52 @@
----
-FIREFOX ADD-ONS (AMO) LISTING - WebBrain v0.9.0
----
-
-NAME: WebBrain β€” Open-Source AI Browser Agent
-
-SUMMARY (up to 250 chars):
-Open-source AI browser agent for Firefox. Use any LLM (OpenAI, Gemini, Anthropic, local models) to read, analyze, and automate web pages via the Firefox sidebar.
-
-DESCRIPTION (up to 8,000 chars):
-WebBrain is an open-source Firefox add-on that turns any LLM into a powerful browser agent β€” integrated seamlessly into your Firefox sidebar.
-
-Unlike proprietary alternatives, WebBrain is provider-agnostic: use OpenAI, Google Gemini, Anthropic Claude, or local models via OpenRouter. Your choice, your API key, your data.
-
-TWO MODES:
-
-πŸ” Ask Mode (Default)
-Read-only page analysis. Ask questions about any web page without modifying anything. Perfect for:
-β€’ Summarizing articles and documents
-β€’ Extracting data from tables and lists
-β€’ Analyzing page structure and content
-β€’ Getting explanations of complex content
-
-πŸ€– Act Mode
-Full browser automation with your permission. Let the AI agent interact with pages on your behalf:
-β€’ Click buttons and links
-β€’ Fill out forms
-β€’ Navigate between pages
-β€’ Scroll and read dynamic content
-β€’ Execute multi-step workflows
-
-FIREFOX INTEGRATION:
-β€’ Native sidebar integration β€” access WebBrain anytime with Ctrl+Shift+E (customizable)
-β€’ Works with Firefox's sidebar_action API
-β€’ Persistent sidebar across tabs and windows
-β€’ Full keyboard shortcuts support
-β€’ Compatible with Firefox 90+
-
-SUPPORTED PROVIDERS:
-β€’ OpenAI (GPT-4o, GPT-4.1, etc.)
-β€’ Google Gemini (Gemini 2.5 Pro, Flash, etc.)
-β€’ Anthropic Claude (Claude Sonnet, Opus, Haiku)
-β€’ OpenRouter (access to 200+ models including local/self-hosted)
-β€’ Any OpenAI-compatible API endpoint
-
-KEY FEATURES:
-β€’ Provider-agnostic β€” switch between AI providers anytime
-β€’ Firefox sidebar UI β€” always accessible, integrated naturally
-β€’ Ask & Act modes β€” read-only analysis or full automation
-β€’ Configurable agent steps β€” control how many actions the agent takes
-β€’ Per-tab conversations β€” separate chat history for each tab
-β€’ Page inspection indicator β€” see when the agent is analyzing a page
-β€’ Dark mode support
-β€’ Open source (GPL-3.0-or-later for version 33.0.0 and later)
-
-PRIVACY & SECURITY:
-WebBrain runs entirely in your browser. Your API keys are stored locally and never sent to any third-party server. All communication goes directly between your Firefox browser and your chosen AI provider. The full source code is available on GitHub for audit.
+FIREFOX ADD-ONS LISTING - proposed English copy, reviewed 2026-10-01
-Website: https://webbrain.one
-GitHub: https://github.com/webbrain-one/webbrain
-License: GPL-3.0-or-later (version 33.0.0 and later; earlier releases remain MIT-licensed)
-Author: Emre Sokullu
+NAME:
+WebBrain β€” AI Browser Agent
+
+SUMMARY (250 chars max):
+Open-source AI browser agent for Firefox. Chat with pages and PDFs, extract data, and automate tasks from your sidebar with your choice of AI.
+
+DESCRIPTION:
+WebBrain is an open-source AI browser agent for Firefox. Ask questions about pages and PDFs, collect information, and automate web tasks from your Firefox sidebar. Work in the browser you already use, with your existing website sessions and your choice of AI model.
+
+READ, RESEARCH AND GET THINGS DONE
+
+β€’ Chat with pages and PDFs: summarize articles, understand documents, translate passages, and ask follow-up questions without copying everything into another app.
+β€’ Automate browser tasks: describe what you need in plain language. The agent can click, type, fill forms, navigate between pages, and work through multiple steps.
+β€’ Extract useful data: collect tables, lists, links and page details for research, comparisons and reports.
+β€’ Schedule tasks and watch pages: run a task later or at a recurring interval, and monitor a page for a condition you specify. Your browser must be running for scheduled work.
+β€’ Reuse your workflows: save successful runs, teach a workflow by demonstrating actions, and import or export workflows to use again.
+
+PRACTICAL TOOLS IN THE SAME SIDEBAR
+
+β€’ Page screenshots: capture the visible part of a tab and save the image.
+β€’ Bulk file downloads: download multiple linked files in one task, including documents and images.
+β€’ Article and video summaries: read New York Times articles and summarize YouTube transcripts through the included reading tools. Availability depends on the source and connected service.
+β€’ CAPTCHA assistance: connect a supported solving service to help continue tasks blocked by supported challenges. Optional setup and a separate service account are required; charges may apply.
+
+CHOOSE WHERE YOUR AI RUNS
+
+Start with WebBrain's built-in model without an API key. For local inference, connect to a model running on your computer through LM Studio or Ollama. For cloud models, connect your own API key for OpenAI, Anthropic Claude, Gemini or OpenRouter. Additional connections are available in Settings.
-CATEGORY: Productivity
+STAY IN CONTROL
+
+Use Ask mode for read-only questions and Act mode for browser actions. Review a plan before execution with Plan before Act, manage site permissions, and stop a run at any time. Dev mode adds page inspection and debugging tools.
+
+Your selected AI service processes the context needed for your request. Local models can keep model inference on your computer; connected reading and CAPTCHA tools use online services. The built-in model's Help Improve WebBrain setting is on by default and can be turned off. See the privacy policy for data-use details.
+
+OFFLINE READING, TOO
+
+Optional Apocalypse Mode adds offline AI, a Wikipedia reader and a reference library. Download the models and archives first; on-device AI requires compatible WebGPU support. Download sizes vary by model and archive.
+
+The extension is free to install. AI and connected services may have separate usage limits or charges. Source code is available under GPL-3.0-or-later.
+
+Open the sidebar and try: "Summarize this page and list the key facts."
+
+Website: https://webbrain.one
+Privacy: https://webbrain.one/privacy
+Source and support: https://github.com/webbrain-one/webbrain
-LANGUAGE: English
+END DESCRIPTION
-PERMISSIONS EXPLAINED:
-β€’ Access to all web pages: Required to read and analyze page content
-β€’ Tabs API: Required to track active tab and maintain per-tab conversation history
-β€’ Storage: Required to save your API keys and settings locally
-β€’ Scripting: Required for Act Mode to interact with web pages on your behalf
+EDITORIAL NOTE - not part of the store description:
+This is proposed copy, not a published store update. The title and summary are separate metadata; Chrome also requires manifest changes in a future package. See store-listing-review.txt for evidence and publication notes. Full-page capture is not yet wired through the Firefox BiDi companion in this checkout. Correct the separate AMO license field to GPL-3.0-or-later as well.
diff --git a/assets/store-listing-review.txt b/assets/store-listing-review.txt
new file mode 100644
index 000000000..1d89b2a34
--- /dev/null
+++ b/assets/store-listing-review.txt
@@ -0,0 +1,76 @@
+WebBrain store listing review - 2026-10-01
+
+Recommendation
+
+Lead with an open-source agent that reads and acts in the user's existing browser. Put concrete outcomes first: understand pages and PDFs, do browser tasks, collect data, schedule work, and repeat workflows. Keep model choice visible as a meaningful setup decision. Group downloads, screenshots, article reading and CAPTCHA assistance beneath the core features. Keep Apocalypse Mode as a distinctive secondary benefit.
+
+Proposed title: WebBrain β€” AI Browser Agent
+Open source remains in the summary and opening sentence. This is a positioning recommendation, not a measured search-ranking or conversion improvement.
+
+Live pages inspected directly in the browser
+
+Chrome: https://chromewebstore.google.com/detail/webbrain/ljhijonmfahplgbbacgcfnaihbjljhhb
+Displayed version: 36.1.0; updated September 15, 2026.
+Firefox: https://addons.mozilla.org/en-US/firefox/addon/webbrain/
+Displayed version: 36.8.0; updated September 20, 2026.
+Local source: a34843d22; both manifests report 38.0.1.
+
+The public descriptions are nearly identical. They omit scheduling, workflows, explicit PDF analysis, bulk downloads, screenshots, article/transcript reading, CAPTCHA integration and Dev mode. The Firefox opening incorrectly says Chrome. Both descriptions say MIT; Firefox's separate license metadata also says MIT. LICENSE specifies GPL-3.0-or-later for 33.0.0 onward.
+
+Offline features occupy a large share of the existing copy. The claims of zero data leakage and unconditional offline continuity are too broad. The blanket WebGPU requirement is also wrong for local server connections. Replace these with specific data-flow and setup statements.
+
+The previous assets/store-listing-*.txt files were labeled v0.9.0 and differed from the live pages. These files now contain proposed copy. No manifest, package, store dashboard, privacy field, license metadata or live listing was changed.
+
+The Chrome rejection
+
+The user supplied the Keyword Spam rejection, reference Yellow Argon. Google specifically flagged the sentence naming LM Studio, Ollama, llama.cpp, OpenAI, Anthropic Claude, Gemini and OpenRouter. This points to the brand-heavy clause; it does not establish a numerical limit on provider names or a complaint about the word provider alone.
+
+The proposed copy removes llama.cpp from the examples: seven external AI connection names become six. It retains two local and four cloud examples, each once in the description, in sentences explaining where inference runs and how the user connects. This changes the listing, not product compatibility. The full catalog belongs in Settings and documentation.
+
+Dropping one name and rewriting the clause cannot guarantee approval. If the reviewer still considers six examples excessive, replace the entire model-choice paragraph with this more conservative option:
+
+Start with WebBrain's built-in model without an API key. For local inference, connect to a model running on your computer through LM Studio or Ollama. You can also connect your own API key for a supported cloud model. Find the available connections in Settings.
+
+This is a replacement option, not an additional paragraph. No appeal or resubmission was made.
+
+Source evidence and claim boundaries
+
+β€’ Core modes/actions: README.md, docs/agent-tools.md, src/{chrome,firefox}/src/agent/tools.js. Ask reads; Act interacts; Dev adds inspection/debugging. Avoid guaranteed completion on every site.
+β€’ PDFs: read_pdf in both agent/tools.js files, agent/pdf-tools.js, ui/pdf-handler.js. Viewer/OCR work appears under 35.0.0 and 36.0.1 in CHANGELOG.md. The copy promises PDF questions/analysis, not perfect extraction from every scanned document.
+β€’ Bulk downloads: download_files in both agent/tools.js files accepts multiple URLs, with 50 total and three concurrent transfers. Avoid unlimited downloader claims.
+β€’ New York Times reading and YouTube transcripts: src/{chrome,firefox}/skills/freeskillz-xyz.md, agent/skills.js, docs/skills.md, docs/privacy-and-data-flow.md. The service-backed skill is enabled by default and removable. NYTimes support entered history in July 2026 (834e4c524, 0b68a52bf). Do not imply publisher affiliation, universal article access or a universal paywall bypass. Source/service availability is explicit in the draft.
+β€’ Scheduling/watches: schedule_task in both agent/tools.js files, agent/scheduler.js, docs/slash-commands.md. One-shot and fixed-interval recurrence exist; calendar/cron recurrence does not. Browser alarms require the browser to run. Do not promise execution with the browser/computer shut down.
+β€’ Workflows/teaching: agent/workflows.js, agent/teacher-mode.js, content/teacher-capture.js, docs/export-and-workflow-formats.md. Teacher mode is documented in 27.1.0. Avoid claiming every demonstrated action can be captured/replayed.
+β€’ Chrome full-page screenshots: agent/agent.js captureFullPageScreenshotForUser and background.js capture_full_page_screenshot. Firefox ui/sidepanel.js:712 still marks --full-page unsupported; its background only routes viewport capture. Shared retired tool names and rendering helpers do not prove a working capture path.
+β€’ Firefox BiDi: firefox-companion/session.mjs:179 allows navigate, click, hover, type, field, key and upload, with no screenshot operation. src/firefox/src/bidi/client.js does not route screenshot capture. The protocol supports full-document screenshots, but this checkout has not wired that capability into WebBrain. Keep viewport wording until the feature is implemented and verified. This is an implementation gap, not an inherent Firefox limitation.
+β€’ CAPTCHA: agent/captcha-solver.js, ui/captcha-settings.js, docs/captcha-provider-coverage.md. Opt-in integration dates to 7.3.0. The expanded seven-service catalog is newer than the displayed store releases. Advertise generic optional assistance, not that new catalog, particular challenge families or guaranteed acceptance. Disclose separate setup/account/charges.
+β€’ Offline mode: docs/apocalypse-mode.md and the live listings. Keep prior-download and WebGPU qualifications attached to on-device inference, not all local model servers.
+β€’ Model setup: README.md and docs/providers-and-models.md. The built-in default requires no API key. Compatibility depends on the selected model and setup; avoid any-LLM guarantees and rapidly aging model-version lists.
+β€’ Privacy: docs/privacy-and-data-flow.md. Relevant context goes to the selected AI service; local inference and online skill calls are distinct. Help Improve WebBrain is on by default for the built-in model and can be disabled. Avoid all-data-stays-local and API-keys-never-leave claims.
+
+This was a source and listing review, not a live end-to-end test of paid solvers or reading services. Store versions lag the checkout. The draft avoids newly expanded solver counts. Verify package-specific behavior when preparing a submission; source presence alone is not proof of a published feature.
+
+Search and conversion
+
+Use AI browser agent, PDFs, data extraction, scheduled tasks, workflows, bulk downloads and the platform-appropriate screenshot phrase within real feature explanations. Do not append keyword, competitor, model-version or website lists. Mention the named reading sources once as use cases.
+
+Suggested screenshot story: the browser agent completing a clear request; PDF questions; a multi-step task; a scheduled watch or reusable workflow; then downloads/screenshots. Use actual UI. Offline mode can follow the core story. No new images were created in this review.
+
+Firefox's current categories are Search Tools and Tabs, and its tags emphasize sites and downloads. Reassess the actual category/tag choices in the dashboard against the core productivity/research use case. Do not add unrelated tags for reach.
+
+Publication notes
+
+Copy only the text between DESCRIPTION and END DESCRIPTION into the long-description field. Title and summary are separate metadata. Chrome derives them from the manifest, so changing those requires a future package. AMO's license field needs a separate correction. The Chrome privacy declaration also merits reconciliation with the built-in service's documented data use before the next submission.
+
+Reference guidance
+
+Chrome listing requirements, including brand lists without substantial added value:
+https://developer.chrome.com/docs/webstore/program-policies/listing-requirements
+Chrome listing guidance and 132-character summary:
+https://developer.chrome.com/docs/webstore/best-listing
+Firefox listing guidance and 250-character summary:
+https://extensionworkshop.com/documentation/develop/create-an-appealing-listing/
+BiDi full-document capture:
+https://developer.mozilla.org/en-US/docs/Web/WebDriver/Reference/BiDi/Modules/browsingContext/captureScreenshot
+
+These sources favor concise, accurate descriptions and natural keywords. They do not establish that any particular brand count guarantees approval or higher ranking.