Open-source AI browser agent for Chrome and Firefox (monorepo) 🧠
JavaScript ai-agentai-agentsai-toolsbrowser-aibrowser-automationbrowser-extensionchrome-aichrome-devtoolschrome-extensionchrome-extensionsfirefox-addonfirefox-addonsfirefox-extensionfirefox-extensionsgemmagemma4minimaxoffline-aioffline-llmqwen
docs(web): name Compass VLD Tiny v5 as the in-browser default
Claudefd948785a7c8
diff --git a/test/run.js b/test/run.jsindex 8ec1a3649..d6d12324b 100644--- a/test/run.js+++ b/test/run.js@@ -43776,8 +43776,9 @@ test('public Apocalypse Mode guide and launch essay document the offline boundar'docs: the guide should distinguish preparation downloads from offline operation');assert.match(guide, /Chromium, with compatible WebGPU hardware[\s\S]*?Firefox[\s\S]*?Chromium-only WebGPU text and vision download stack is not shown/,'docs: the guide should state the browser-specific WebGPU boundary');- assert.match(guide, /LiquidAI\/LFM2\.5-2\.6B-ONNX[\s\S]*?does not silently change the provider/,- 'docs: the local text model should be named without implying global provider selection');+ assert.match(guide, /webbrain-one\/webbrain-compass-vld-tiny-v5-onnx[\s\S]*?does not silently change the provider/,+ 'docs: the default local text model should be named without implying global provider selection');+ assert.match(chineseGuide, /webbrain-one\/webbrain-compass-vld-tiny-v5-onnx/, 'docs: the Chinese guide names the same default text model');assert.match(guide, /about 810 MB[\s\S]*?WebBrain VL 2 450M/,'docs: the shipped local vision model and approximate download size should be current');assert.match(guide, /Wikipedia reader[\s\S]*?Emergency Box[\s\S]*?Medical guidance becomes outdated/,diff --git a/web/docs/apocalypse-mode/index.html b/web/docs/apocalypse-mode/index.htmlindex 5e065c9d5..6da6c4f7f 100644--- a/web/docs/apocalypse-mode/index.html+++ b/web/docs/apocalypse-mode/index.html@@ -19,15 +19,15 @@<aside class="docs-sidebar" aria-label="Documentation"><div class="sidebar-group"><div class="sidebar-label">Start here</div><nav><a href="/docs/">Overview</a><a href="/docs/faq/">FAQ</a></nav></div><div class="sidebar-group"><div class="sidebar-label">Use WebBrain</div><nav><a href="/docs/settings/">Settings, explained</a><a href="/docs/providers/">Providers & models</a><a href="/docs/easy-cli-proxy/">Subscription proxy</a><a href="/docs/safety/">Modes, safety & privacy</a><a href="/docs/apocalypse-mode/" aria-current="page">Apocalypse Mode</a><a href="/docs/formats/">Exports & workflows</a><a href="/docs/mcp/">MCP server</a><a href="/docs/lm-studio/">LM Studio plugin</a><a href="/docs/ollama/">Ollama</a></nav></div><div class="sidebar-group"><div class="sidebar-label">For builders</div><nav><a href="https://github.com/webbrain-one/webbrain">Source code ↗</a></nav></div><div class="sidebar-group"><div class="sidebar-label">Community</div><nav><a href="https://discord.gg/cgC325ssfw">Discord Support</a></nav></div></aside><main class="doc-main" id="main"><div class="breadcrumb"><a href="/docs/">Docs</a> / Apocalypse Mode</div>- <section class="doc-hero"><div class="eyebrow">WebBrain, ready when the internet isn’t</div><h1>Prepare local intelligence and practical knowledge before you need them.</h1><p class="lede">Apocalypse Mode assembles an offline-ready kit inside your browser: on-device text and vision models on supported Chromium systems, a searchable Wikipedia archive, and an Emergency Box of books and field references.</p><div class="mode-strip" aria-label="The three essentials in the Chromium Apocalypse Mode setup"><div class="mode ask"><strong>Text model</strong><span>1.55 GB · local WebGPU inference</span></div><div class="mode act"><strong>Vision model</strong><span>optional · screenshots stay local</span></div><div class="mode dev"><strong>Wikipedia</strong><span>Simple English · searchable offline</span></div></div></section>+ <section class="doc-hero"><div class="eyebrow">WebBrain, ready when the internet isn’t</div><h1>Prepare local intelligence and practical knowledge before you need them.</h1><p class="lede">Apocalypse Mode assembles an offline-ready kit inside your browser: on-device text and vision models on supported Chromium systems, a searchable Wikipedia archive, and an Emergency Box of books and field references.</p><div class="mode-strip" aria-label="The three essentials in the Chromium Apocalypse Mode setup"><div class="mode ask"><strong>Text model</strong><span>2.2 GB · local WebGPU inference</span></div><div class="mode act"><strong>Vision model</strong><span>optional · screenshots stay local</span></div><div class="mode dev"><strong>Wikipedia</strong><span>Simple English · searchable offline</span></div></div></section><section class="doc-section" id="meaning"><h2>What “offline” means</h2><p>You need an internet connection for the first download. Once the model files, Wikipedia archive, and chosen documents are stored, their core inference, search, and reading paths run locally from the browser profile. WebGPU performs model inference on the device; downloaded screenshots do not need a remote vision service.</p><div class="callout tip"><strong class="callout-title">Prepare on an ordinary day</strong><p>Large downloads and storage checks are much easier while power, connectivity, and time are predictable. Enable the mode, let every required item reach <strong>Ready</strong>, and test the readers before treating the kit as available.</p></div><p>Apocalypse Mode does not make every website or WebBrain task work offline. Live pages, hosted model providers, current news, and any resource you did not download still require a network. It prepares a self-contained local core; it is not a copy of the internet.</p></section><section class="doc-section" id="browsers"><h2>The Chromium and Firefox builds are different</h2><div class="table-wrap"><table><thead><tr><th>Build</th><th>What enabling the mode prepares</th><th>Ready and Emergency Box</th></tr></thead><tbody><tr><td>Chromium, with compatible WebGPU hardware</td><td>Starts the local text model and the newest Simple English text-only Wikipedia archive. The vision model is optional and downloads only when you click its Download button. Model transfers share one local worker and run in sequence.</td><td><strong>Ready</strong> requires the text model and the offline library; the vision model is optional. The Emergency Box unlocks when they are complete.</td></tr><tr><td>Firefox</td><td>Provides the offline Wikipedia library, reader, and Emergency Box. The Chromium-only WebGPU text and vision download stack is not shown.</td><td>Archive and document controls remain available without pretending Firefox has the Chromium WebGPU runtime.</td></tr></tbody></table></div><p>If the text and vision cards do not appear, first confirm that you are using the Chromium build on hardware and a browser version with WebGPU support.</p></section>- <section class="doc-section" id="setup"><h2>Build the basic kit</h2><ol class="step-list"><li><strong>Open Apocalypse Mode.</strong> Use the nuclear-symbol Apocalypse Mode control in the Settings header.</li><li><strong>Check your storage and connection.</strong> The Chromium basics include a 1.55 GB text model and the current Simple English Wikipedia archive. An in-browser vision model is optional: about 1.45 GB for Ternary Bonsai 2 2B Vision or about 810 MB for WebBrain VL 2 450M. Leave additional room for temporary download state and Emergency Box resources.</li><li><strong>Turn the mode on.</strong> On supported Chromium systems, WebBrain starts the text model and the required Wikipedia archive automatically. It never downloads a vision model on its own.</li><li><strong>Keep the browser open for model downloads.</strong> You may switch tabs or close Settings, but Chrome must stay open while the local model transfer continues. Follow the per-item progress instead of relying only on the total storage number.</li><li><strong>Resolve anything not ready.</strong> Resume a paused item, retry an error, or free storage. Removing a completed model deletes its cached files and requires another download before offline use.</li><li><strong>Test the kit.</strong> Use the vision test, open the Wikipedia reader, search for an article, and open at least one installed PDF. A download is not a plan until you have proved you can read it.</li></ol></section>+ <section class="doc-section" id="setup"><h2>Build the basic kit</h2><ol class="step-list"><li><strong>Open Apocalypse Mode.</strong> Use the nuclear-symbol Apocalypse Mode control in the Settings header.</li><li><strong>Check your storage and connection.</strong> The Chromium basics include a 2.2 GB text model that also reads images and the current Simple English Wikipedia archive. A separate in-browser vision model is optional: about 1.45 GB for Ternary Bonsai 2 2B Vision or about 810 MB for WebBrain VL 2 450M. Leave additional room for temporary download state and Emergency Box resources.</li><li><strong>Turn the mode on.</strong> On supported Chromium systems, WebBrain starts the text model and the required Wikipedia archive automatically. It never downloads a vision model on its own.</li><li><strong>Keep the browser open for model downloads.</strong> You may switch tabs or close Settings, but Chrome must stay open while the local model transfer continues. Follow the per-item progress instead of relying only on the total storage number.</li><li><strong>Resolve anything not ready.</strong> Resume a paused item, retry an error, or free storage. Removing a completed model deletes its cached files and requires another download before offline use.</li><li><strong>Test the kit.</strong> Use the vision test, open the Wikipedia reader, search for an article, and open at least one installed PDF. A download is not a plan until you have proved you can read it.</li></ol></section>- <section class="doc-section" id="models"><h2>What the local models do</h2><div class="card-grid"><div class="card"><span class="tab-key">Text · 1.55 GB</span><h3>LFM2.5 2.6B</h3><p>WebBrain downloads <code>LiquidAI/LFM2.5-2.6B-ONNX</code> in <code>q4f16</code> form for local text generation through WebGPU. Use the nuclear control in standalone chat after it is ready.</p></div><div class="card"><span class="tab-key">Vision · optional</span><h3>Ternary Bonsai 2 2B Vision or WebBrain VL 2 450M</h3><p>The local vision layer interprets screenshots on the device, turning visual evidence into context without sending the image to a remote vision provider. The Vision Model card shares one choice with <strong>Providers → Vision</strong>: <strong>Auto</strong> (default) uses Ternary Bonsai 2 2B Vision (about 1.45 GB) once it is downloaded; you can also pick WebBrain VL 2 450M (about 810 MB, vision only, needs consent on the Providers page first) or Off. If your Text Model is Ternary Bonsai 2 2B Vision, it already reads images and needs no extra download.</p></div></div><div class="callout warning"><strong class="callout-title">Apocalypse Mode does not replace your selected provider</strong><p>The downloaded text model is a run-scoped option for standalone chat. Enabling or downloading it does not silently change the provider used by ordinary WebBrain conversations.</p></div></section>+ <section class="doc-section" id="models"><h2>What the local models do</h2><div class="card-grid"><div class="card"><span class="tab-key">Text · 2.2 GB</span><h3>Compass VLD Tiny v5</h3><p>WebBrain downloads <code>webbrain-one/webbrain-compass-vld-tiny-v5-onnx</code>, its fine-tune of MiniCPM-V-4.7-1B, for local chat with tool calls through WebGPU. The same model reads screenshots and answers WebBrain’s decision questions. Use the nuclear control in standalone chat after it is ready.</p></div><div class="card"><span class="tab-key">Vision · optional</span><h3>Ternary Bonsai 2 2B Vision or WebBrain VL 2 450M</h3><p>The local vision layer interprets screenshots on the device, turning visual evidence into context without sending the image to a remote vision provider. The Vision Model card shares one choice with <strong>Providers → Vision</strong>: <strong>Auto</strong> (default) uses Compass VLD Tiny v5 once it is downloaded, and Ternary Bonsai 2 2B Vision (about 1.45 GB) when only that one is; you can also pick WebBrain VL 2 450M (about 810 MB, vision only, needs consent on the Providers page first) or Off. If your Text Model is Compass VLD Tiny v5 or Ternary Bonsai 2 2B Vision, it already reads images and needs no extra download.</p></div></div><div class="callout warning"><strong class="callout-title">Apocalypse Mode does not replace your selected provider</strong><p>The downloaded text model is a run-scoped option for standalone chat. Enabling or downloading it does not silently change the provider used by ordinary WebBrain conversations.</p></div></section><section class="doc-section" id="wikipedia"><h2>Keep Wikipedia you can actually search</h2><p>The basic setup downloads the newest compact, text-only Simple English archive. Open <strong>Manage archives</strong> when you want another language, a fuller edition, or an edition with images. WebBrain resolves the archive’s current version, date, and exact size before installation.</p><div class="table-wrap"><table><thead><tr><th>Choice</th><th>Trade-off</th><th>Good for</th></tr></thead><tbody><tr><td>Text only</td><td>Smaller and faster to download; article text without the full image payload.</td><td>Compact reference, local factual retrieval, and broad language coverage.</td></tr><tr><td>With images</td><td>Uses substantially more storage.</td><td>Articles where diagrams, maps, and photographs matter.</td></tr><tr><td>Import a <code>.zim</code> file</td><td>You manage the source file and may need to reauthorize access after a restart.</td><td>Existing Kiwix collections or archives moved between machines.</td></tr></tbody></table></div><p>The Wikipedia reader searches article titles and opens the matching article from local storage. Local standalone chat can retrieve compact, attributed passages from the installed archive. When you replace the required Simple English archive, WebBrain keeps the old copy until the replacement has downloaded and verified successfully.</p></section>diff --git a/web/docs/providers/index.html b/web/docs/providers/index.htmlindex d80472ec8..5454dea64 100644--- a/web/docs/providers/index.html+++ b/web/docs/providers/index.html@@ -35,7 +35,7 @@<section class="doc-section" id="troubleshoot"><h2>When Test Connection fails</h2><ol><li><strong>Connection refused:</strong> start the local server and confirm the port.</li><li><strong>CORS or origin error:</strong> allow the WebBrain extension origin in the local server.</li><li><strong>Unauthorized:</strong> paste a current key for the same provider as the base URL.</li><li><strong>Model not found:</strong> copy the exact model ID from the provider or use Load models.</li><li><strong>Requests time out:</strong> raise the LLM request timeout in Settings → Budgets for large local models on CPU.</li><li><strong>Runs fail after a few turns:</strong> confirm the real context window; lower the prompt tier or load the model with a larger context.</li></ol><div class="callout tip"><strong class="callout-title">Test before you switch</strong><p>Keep the working provider active while you configure a new one. Save and test the new card first; only then click Set Active.</p></div></section>- <section class="doc-section" id="assistive-models"><h2>Optional assistive models</h2><p>The active provider handles the main conversation, planning and final replies. The other tabs of the Providers page hold a separate <strong>Vision</strong> model, a <strong>Speech</strong> (speech-to-text) endpoint, <strong>Generative</strong> media, <strong>Decision</strong> models and <strong>SafeSocial</strong>. These settings are independent of provider filtering and selection.</p><p>Vision and Decision models default to <strong>Auto</strong>: in Chrome they use Ternary Bonsai 2 2B Vision once it is downloaded (or Compass’s managed decision model for decisions while WebBrain Cloud is active), and nothing before that. A decision model only answers yes-or-no and multiple-choice questions such as “Is this task finished?”; turn it <strong>On</strong> to use OpenRouter, TypeSafe, a local server, D1 or the in-browser model for every decision, or <strong>Off</strong> to never use one. Existing permissions and completion checks still apply. See <a href="/docs/settings/#multimodal">Vision, speech and decision models</a> for setup and data-sharing details.</p></section>+ <section class="doc-section" id="assistive-models"><h2>Optional assistive models</h2><p>The active provider handles the main conversation, planning and final replies. The other tabs of the Providers page hold a separate <strong>Vision</strong> model, a <strong>Speech</strong> (speech-to-text) endpoint, <strong>Generative</strong> media, <strong>Decision</strong> models and <strong>SafeSocial</strong>. These settings are independent of provider filtering and selection.</p><p>Vision and Decision models default to <strong>Auto</strong>: in Chrome they use Compass VLD Tiny v5 once it is downloaded, and nothing before that. Without it, a downloaded Ternary Bonsai 2 2B Vision reads screenshots and a downloaded D1 answers decisions; while WebBrain Cloud is active, Compass’s managed decision model answers decisions until one of these is downloaded. A decision model only answers yes-or-no and multiple-choice questions such as “Is this task finished?”; turn it <strong>On</strong> to use OpenRouter, TypeSafe, a local server, D1 or the in-browser model for every decision, or <strong>Off</strong> to never use one. Existing permissions and completion checks still apply. See <a href="/docs/settings/#multimodal">Vision, speech and decision models</a> for setup and data-sharing details.</p></section><nav class="next-links" aria-label="Next guides"><a href="/docs/settings/"><span>Back</span>← Settings, explained</a><a href="/docs/safety/"><span>Next</span>Modes, safety & privacy →</a></nav></main><aside class="on-this-page" aria-label="On this page"><h2>On this page</h2><a href="#quick-choice">Quick choice</a><a href="#catalog">Provider catalog</a><a href="#setup">Configure</a><a href="#fields">Fields</a><a href="#local">Local models</a><a href="#tiers">Prompt tiers</a><a href="#troubleshoot">Troubleshoot</a></aside>diff --git a/web/docs/settings/index.html b/web/docs/settings/index.htmlindex f77c2a14d..a4fd394c0 100644--- a/web/docs/settings/index.html+++ b/web/docs/settings/index.html@@ -24,7 +24,7 @@<div class="breadcrumb"><a href="/docs/">Docs</a> / Settings</div><section class="doc-hero"><div class="eyebrow">Settings, without guesswork</div><h1>Know what to change—and what to leave alone.</h1><p class="lede">Open Settings from the gear in the side panel or from your browser’s extension options. The chat model and its helper models have their own <strong>Providers</strong> page: open it from <strong>More…</strong> in the side panel’s model menu, or from the Providers link at the top of Settings. Site permissions have a page too: open it with <strong>Manage site permissions</strong> in the <strong>⚠</strong> menu next to the microphone. Most people only need General, Providers and Permissions; the other tabs solve specific problems.</p><div class="meta">Screenshots from the Settings reorganization · synthetic configuration, no real credentials</div></section>- <section class="doc-section" id="defaults"><h2>A good default setup</h2><div class="table-wrap"><table><thead><tr><th>Control</th><th>Recommended</th><th>Why</th></tr></thead><tbody><tr><td>Plan before Act</td><td>Try planning</td><td>Uses a plan when available and safely falls back to a read-only turn if planner formatting fails.</td></tr><tr><td>Plan review stops</td><td>Below confidence threshold</td><td>Low-confidence plans pause; confident plans can continue.</td></tr><tr><td>Review below confidence</td><td>90%</td><td>A conservative balance between supervision and flow.</td></tr><tr><td>Max Agent Steps</td><td>130</td><td>Enough room for multi-step work. Stop the run yourself if behavior looks repetitive.</td></tr><tr><td>LLM request timeout</td><td>120 seconds</td><td>Cloud models usually answer sooner; large local models may need the extra startup time.</td></tr><tr><td>Auto-screenshot</td><td>On navigation</td><td>One picture after the agent opens a page, not after every click or keystroke. Choose On state change when the vision model should check each action.</td></tr><tr><td>Cloud cost allowances</td><td>$10 per session and $10 in total</td><td>A mechanical stop for paid model calls; raise or lower it in Budgets.</td></tr><tr><td>Scheduled action confirmation</td><td>On</td><td>Unattended work still pauses before consequential actions.</td></tr><tr><td>Permission gate</td><td>On</td><td>WebBrain asks before it first acts on a site and remembers your explicit choice. The ⚠ button next to the microphone in the side panel switches it too.</td></tr><tr><td>Decision models</td><td>Auto</td><td>Nothing to set up. Uses Ternary Bonsai 2 2B Vision for completion checks once it is downloaded (or Compass’s managed decision model while WebBrain Cloud is active), and no decision model otherwise.</td></tr><tr><td>In-browser vision model</td><td>Auto</td><td>Reads screenshots with Ternary Bonsai 2 2B Vision once it is downloaded, only when your chat model cannot see images.</td></tr></tbody></table></div></section>+ <section class="doc-section" id="defaults"><h2>A good default setup</h2><div class="table-wrap"><table><thead><tr><th>Control</th><th>Recommended</th><th>Why</th></tr></thead><tbody><tr><td>Plan before Act</td><td>Try planning</td><td>Uses a plan when available and safely falls back to a read-only turn if planner formatting fails.</td></tr><tr><td>Plan review stops</td><td>Below confidence threshold</td><td>Low-confidence plans pause; confident plans can continue.</td></tr><tr><td>Review below confidence</td><td>90%</td><td>A conservative balance between supervision and flow.</td></tr><tr><td>Max Agent Steps</td><td>130</td><td>Enough room for multi-step work. Stop the run yourself if behavior looks repetitive.</td></tr><tr><td>LLM request timeout</td><td>120 seconds</td><td>Cloud models usually answer sooner; large local models may need the extra startup time.</td></tr><tr><td>Auto-screenshot</td><td>On navigation</td><td>One picture after the agent opens a page, not after every click or keystroke. Choose On state change when the vision model should check each action.</td></tr><tr><td>Cloud cost allowances</td><td>$10 per session and $10 in total</td><td>A mechanical stop for paid model calls; raise or lower it in Budgets.</td></tr><tr><td>Scheduled action confirmation</td><td>On</td><td>Unattended work still pauses before consequential actions.</td></tr><tr><td>Permission gate</td><td>On</td><td>WebBrain asks before it first acts on a site and remembers your explicit choice. The ⚠ button next to the microphone in the side panel switches it too.</td></tr><tr><td>Decision models</td><td>Auto</td><td>Nothing to set up. Uses Compass VLD Tiny v5 for completion checks once it is downloaded, or D1 when only D1 is (Compass’s managed decision model while WebBrain Cloud is active and neither is), and no decision model otherwise.</td></tr><tr><td>In-browser vision model</td><td>Auto</td><td>Reads screenshots with Compass VLD Tiny v5 once it is downloaded (Ternary Bonsai 2 2B Vision when only that one is), only when your chat model cannot see images.</td></tr></tbody></table></div></section><section class="doc-section" id="general"><div class="eyebrow">Everyday</div><h2>General: how the agent behaves</h2><p>Appearance and download location are cosmetic. The controls below them change the agent loop: how it plans, when you review, and what it captures. Language and zoom are picked in the side panel, so they are not repeated here.</p><figure class="shot"><a href="/docs/assets/screenshots/settings-general.jpg"><img src="/docs/assets/screenshots/settings-general.jpg" alt="WebBrain General settings with the search box, appearance, download folder, planning and plan review controls"></a><figcaption>General starts with the controls you are most likely to use. Click the image to view it at full size.</figcaption></figure><h3>Planning controls</h3><ul><li><strong>Try planning</strong> is the best general choice. WebBrain attempts a structured plan before Act or Dev tools run; if planner formatting still fails after repair, that turn continues with the Ask prompt and read-only tools. When Auto picks Act or Dev for a message and is sure it is a short, direct task, such as opening a page, searching, or filling in and submitting one form, Try skips the full plan and runs only the compact intent check. Tasks with several goals, many items, several sites, scheduling, or payments, deletions and other risky steps still get a plan, and so does every Act or Dev turn while plan review is set to always ask.</li><li><strong>Strict planning</strong> cancels the run when a valid plan cannot be produced. Use it where “no plan, no action” matters more than convenience.</li><li><strong>Off</strong> removes the extra full-plan call, while retaining the compact intent check and its safe read-only fallback. It can be faster and cheaper, but you lose the plan review checkpoint.</li></ul>@@ -49,11 +49,11 @@<section class="doc-section" id="providers"><div class="eyebrow">Everyday</div><h2>Providers: choose the brain</h2><p>Providers is a page of its own. Open it from <strong>More…</strong> in the side panel’s model menu or from the <strong>Providers</strong> link at the top of Settings. Its first tab, <strong>LLM</strong>, lists the chat providers; the other tabs hold the helper models described below.</p><p>Only one provider is active at a time. Filter the list by Local, Cloud, or Router; search by provider, model, or field; then open a card to edit it. The safe sequence is <strong>Save → Test Connection → Set Active</strong>.</p><figure class="shot"><a href="/docs/assets/screenshots/settings-providers.jpg"><img src="/docs/assets/screenshots/settings-providers.jpg" alt="WebBrain Providers page, LLM tab, with category filters, search and the active provider card"></a><figcaption>The active provider remains visible even when category filters are applied. Example model names are illustrative.</figcaption></figure><p>See <a href="/docs/providers/">Providers & models</a> for local server URLs, context windows, prompt tiers, and troubleshooting.</p></section>- <section class="doc-section" id="multimodal"><div class="eyebrow">When needed</div><h2>Vision, speech and decision models: support the main agent</h2><p>Choose the main conversation model on the <strong>LLM</strong> tab of the Providers page. The page’s other tabs hold optional helper models: <strong>Vision</strong>, <strong>Speech</strong>, <strong>Generative</strong> (images and video), <strong>Decision</strong> and <strong>SafeSocial</strong>. Configuring one does not change your active provider.</p><p>Vision and Decision models both default to <strong>Auto</strong>. In Chrome with hardware WebGPU, Auto uses <strong>Ternary Bonsai 2 2B Vision</strong> (about 1.45 GB, runs on your device) once it is downloaded (while WebBrain Cloud is active, Compass’s managed decision model answers decisions instead), and nothing before that. The <strong>Download a local model</strong> button on the install page, <strong>Providers → WebGPU</strong>, the Vision card and Apocalypse Mode all download that same model, so one download covers chat, screenshots and decisions.</p>- <h3>Vision</h3><p>Pick what reads screenshots. An OpenAI-compatible endpoint, if you save one, reads every screenshot while the active provider keeps planning and tool calls. Without one, the active model reads screenshots itself if it can see images; if it cannot, the in-browser vision model does (Chrome only). Leave everything at its default to let the active provider handle vision. Screenshot limits are in Settings → Budgets and redaction in Settings → Advanced.</p><p>The in-browser vision model has four settings:</p><ul><li><strong>Auto</strong> (default): uses Ternary Bonsai 2 2B Vision once it is downloaded and stays off until then.</li><li><strong>Ternary Bonsai 2 2B Vision</strong>: about 1.45 GB. The same model also chats and answers decisions.</li><li><strong>WebBrain VL 2 450M</strong>: about 810 MB, vision only. Its first download asks for your consent and checks WebGPU.</li><li><strong>Off</strong>: no in-browser vision model.</li></ul><p>Changing the setting never downloads anything; use the card’s <strong>Download</strong> button. Tasks never wait for a download, and screenshots read in the browser stay on your device. Firefox has no in-browser vision model.</p><figure class="shot"><a href="/docs/assets/screenshots/providers-vision.jpg"><img src="/docs/assets/screenshots/providers-vision.jpg" alt="Vision tab of the Providers page: the in-browser vision model set to Auto and the optional OpenAI-compatible endpoint"></a><figcaption>A fresh install: the in-browser vision model is on Auto and stays off until Ternary Bonsai 2 2B Vision is downloaded.</figcaption></figure>+ <section class="doc-section" id="multimodal"><div class="eyebrow">When needed</div><h2>Vision, speech and decision models: support the main agent</h2><p>Choose the main conversation model on the <strong>LLM</strong> tab of the Providers page. The page’s other tabs hold optional helper models: <strong>Vision</strong>, <strong>Speech</strong>, <strong>Generative</strong> (images and video), <strong>Decision</strong> and <strong>SafeSocial</strong>. Configuring one does not change your active provider.</p><p>Vision and Decision models both default to <strong>Auto</strong>. In Chrome with hardware WebGPU, Auto uses <strong>Compass VLD Tiny v5</strong> (about 2.2 GB, runs on your device) once it is downloaded, and nothing before that; while WebBrain Cloud is active, Compass’s managed decision model answers decisions until then. The <strong>Download a local model</strong> button on the install page, <strong>Providers → WebGPU</strong>, the Vision card and Apocalypse Mode all download that same model, so one download covers chat, screenshots and decisions.</p>+ <h3>Vision</h3><p>Pick what reads screenshots. An OpenAI-compatible endpoint, if you save one, reads every screenshot while the active provider keeps planning and tool calls. Without one, the active model reads screenshots itself if it can see images; if it cannot, the in-browser vision model does (Chrome only). Leave everything at its default to let the active provider handle vision. Screenshot limits are in Settings → Budgets and redaction in Settings → Advanced.</p><p>The in-browser vision model has five settings:</p><ul><li><strong>Auto</strong> (default): uses Compass VLD Tiny v5 once it is downloaded, Ternary Bonsai 2 2B Vision when only that one is, and stays off until then.</li><li><strong>Compass VLD Tiny v5</strong>: about 2.2 GB. The same model also chats and answers decisions.</li><li><strong>Ternary Bonsai 2 2B Vision</strong>: about 1.45 GB. The same model also chats.</li><li><strong>WebBrain VL 2 450M</strong>: about 810 MB, vision only. Its first download asks for your consent and checks WebGPU.</li><li><strong>Off</strong>: no in-browser vision model.</li></ul><p>Changing the setting never downloads anything; use the card’s <strong>Download</strong> button. Tasks never wait for a download, and screenshots read in the browser stay on your device. Firefox has no in-browser vision model.</p><figure class="shot"><a href="/docs/assets/screenshots/providers-vision.jpg"><img src="/docs/assets/screenshots/providers-vision.jpg" alt="Vision tab of the Providers page: the in-browser vision model set to Auto and the optional OpenAI-compatible endpoint"></a><figcaption>A fresh install: the in-browser vision model is on Auto and stays off until Compass VLD Tiny v5 is downloaded.</figcaption></figure><h3>Speech</h3><p>Configure a Whisper-compatible endpoint for recordings. Leave it blank and WebBrain tries configured providers in this order: OpenAI, Groq, LM Studio, then llama.cpp.</p><h3>Generative</h3><p>Connect fal.ai, OpenRouter, ComfyRouter or a local ComfyUI server for the image and video generation tool. Nothing is configured by default.</p>- <h3>Decision</h3><p>A decision model is a small, fast model that answers the agent’s yes-or-no and multiple-choice questions, such as “Is this task finished?”, so your chat model doesn’t have to. It never plans, clicks or types. Choose one mode:</p><ul><li><strong>Auto</strong> (default): nothing to set up. In Chrome it uses Ternary Bonsai 2 2B Vision once it is downloaded, to check whether side panel tasks and scheduled watches and tasks really finished. While WebBrain Cloud (Compass) is your chat provider, Compass’s decision model answers instead. Otherwise no decision model is used. The card shows which one is in use.</li><li><strong>On</strong>: uses the provider you set up for every decision: OpenRouter, TypeSafe, a local System One server, D1, or the in-browser chat model in Chrome. All decisions are on by default. Open <strong>Choose which decisions use it</strong> to turn off individual ones (Verify completion, Verify scheduled watches, Verify scheduled task completion, Fast classifications, Fast browser actions — Experimental) and to set thresholds.</li><li><strong>Off</strong>: never uses a decision model, Compass’s included. Your chat model still checks whether tasks are finished.</li></ul><p>A mode change applies at once; On’s provider fields use <strong>Save</strong>. <strong>Test Connection</strong> sends one request with fixed example data. <strong>Reset</strong> returns to Auto and deletes saved decision keys. Cloud decision providers receive limited task and page data and may cost money; Strict Secret Mode disables decision models. Profiles that had decision outsourcing turned on before these modes keep working as On with their previous switches.</p><figure class="shot"><a href="/docs/assets/screenshots/settings-decision-models.jpg"><img src="/docs/assets/screenshots/settings-decision-models.jpg" alt="Decision models settings with Auto selected and the status line showing that no decision model is in use yet"></a><figcaption>Auto on a fresh install: nothing is downloaded yet, so no decision model is in use (unless WebBrain Cloud is your active provider, which uses Compass’s decision model). Provider fields appear only with On.</figcaption></figure>+ <h3>Decision</h3><p>A decision model is a small, fast model that answers the agent’s yes-or-no and multiple-choice questions, such as “Is this task finished?”, so your chat model doesn’t have to. It never plans, clicks or types. Choose one mode:</p><ul><li><strong>Auto</strong> (default): nothing to set up. In Chrome it uses Compass VLD Tiny v5 once it is downloaded, to check whether side panel tasks and scheduled watches and tasks really finished, and D1 when only D1 is downloaded. While WebBrain Cloud (Compass) is your chat provider, Compass’s decision model answers until one of them is downloaded. Otherwise no decision model is used. The card shows which one is in use.</li><li><strong>On</strong>: uses the provider you set up for every decision: OpenRouter, TypeSafe, a local System One server, D1, or the in-browser chat model in Chrome. All decisions are on by default. Open <strong>Choose which decisions use it</strong> to turn off individual ones (Verify completion, Verify scheduled watches, Verify scheduled task completion, Fast classifications, Fast browser actions — Experimental) and to set thresholds.</li><li><strong>Off</strong>: never uses a decision model, Compass’s included. Your chat model still checks whether tasks are finished.</li></ul><p>A mode change applies at once; On’s provider fields use <strong>Save</strong>. <strong>Test Connection</strong> sends one request with fixed example data. <strong>Reset</strong> returns to Auto and deletes saved decision keys. Cloud decision providers receive limited task and page data and may cost money; Strict Secret Mode disables decision models. Profiles that had decision outsourcing turned on before these modes keep working as On with their previous switches.</p><figure class="shot"><a href="/docs/assets/screenshots/settings-decision-models.jpg"><img src="/docs/assets/screenshots/settings-decision-models.jpg" alt="Decision models settings with Auto selected and the status line showing that no decision model is in use yet"></a><figcaption>Auto on a fresh install: nothing is downloaded yet, so no decision model is in use (unless WebBrain Cloud is your active provider, which uses Compass’s decision model). Provider fields appear only with On.</figcaption></figure><h3>SafeSocial</h3><p>An optional on-device image classifier for Instagram feeds. It is off until you turn it on; the model downloads once and runs in the browser.</p></section>diff --git a/web/docs/zh/apocalypse-mode/index.html b/web/docs/zh/apocalypse-mode/index.htmlindex 3f2b5dc12..f63a40ca1 100644--- a/web/docs/zh/apocalypse-mode/index.html+++ b/web/docs/zh/apocalypse-mode/index.html@@ -18,15 +18,15 @@<aside class="docs-sidebar" aria-label="文档目录"><div class="sidebar-group"><div class="sidebar-label">从这里开始</div><nav><a href="/docs/zh/">概览</a><a href="/docs/zh/faq/">常见问题</a></nav></div><div class="sidebar-group"><div class="sidebar-label">使用 WebBrain</div><nav><a href="/docs/zh/settings/">设置说明</a><a href="/docs/zh/providers/">提供商与模型</a><a href="/docs/zh/easy-cli-proxy/">订阅代理</a><a href="/docs/zh/safety/">模式、安全与隐私</a><a href="/docs/zh/apocalypse-mode/" aria-current="page">末日模式</a><a href="/docs/zh/formats/">导出与工作流</a></nav></div><div class="sidebar-group"><div class="sidebar-label">面向开发者</div><nav><a href="https://github.com/webbrain-one/webbrain">源代码 ↗</a></nav></div><div class="sidebar-group"><div class="sidebar-label">社区</div><nav><a href="https://discord.gg/cgC325ssfw">Discord 支持</a></nav></div></aside><main class="doc-main" id="main"><div class="breadcrumb"><a href="/docs/zh/">文档</a> / 末日模式</div>- <section class="doc-hero"><div class="eyebrow">WebBrain,当互联网不可用时也能使用</div><h1>提前准备好本地智能和实用知识。</h1><p class="lede">末日模式在你的浏览器中组装一套离线就绪的工具包:在支持的 Chromium 系统上运行设备端文本和视觉模型,可搜索的维基百科存档,以及包含书籍和现场参考资料的应急箱。</p><div class="mode-strip" aria-label="Chromium 末日模式设置中的三个核心要素"><div class="mode ask"><strong>文本模型</strong><span>1.55 GB · 本地 WebGPU 推理</span></div><div class="mode act"><strong>视觉模型</strong><span>可选 · 截图保留在本地</span></div><div class="mode dev"><strong>维基百科</strong><span>简易英语版 · 可离线搜索</span></div></div></section>+ <section class="doc-hero"><div class="eyebrow">WebBrain,当互联网不可用时也能使用</div><h1>提前准备好本地智能和实用知识。</h1><p class="lede">末日模式在你的浏览器中组装一套离线就绪的工具包:在支持的 Chromium 系统上运行设备端文本和视觉模型,可搜索的维基百科存档,以及包含书籍和现场参考资料的应急箱。</p><div class="mode-strip" aria-label="Chromium 末日模式设置中的三个核心要素"><div class="mode ask"><strong>文本模型</strong><span>2.2 GB · 本地 WebGPU 推理</span></div><div class="mode act"><strong>视觉模型</strong><span>可选 · 截图保留在本地</span></div><div class="mode dev"><strong>维基百科</strong><span>简易英语版 · 可离线搜索</span></div></div></section><section class="doc-section" id="meaning"><h2>"离线"的含义</h2><p>首次下载需要互联网连接。模型文件、维基百科存档和所选文档存储后,其核心推理、搜索和阅读路径从浏览器配置文件本地运行。WebGPU 在设备上执行模型推理;下载的截图不需要远程视觉服务。</p><div class="callout tip"><strong class="callout-title">在平常的日子里准备</strong><p>在电力、网络和时间都可预测时进行大文件下载和存储检查要容易得多。启用模式,让每个必要项目达到<strong>就绪</strong>状态,并在将工具包视为可用之前测试阅读器。</p></div><p>末日模式不会让每个网站或 WebBrain 任务都能离线工作。实时页面、托管模型提供商、当前新闻以及你未下载的任何资源仍需要网络。它准备了一个自包含的本地核心;它不是互联网的副本。</p></section><section class="doc-section" id="browsers"><h2>Chromium 和 Firefox 版本有所不同</h2><div class="table-wrap"><table><thead><tr><th>版本</th><th>启用模式后准备的内容</th><th>就绪与应急箱</th></tr></thead><tbody><tr><td>Chromium,具有兼容的 WebGPU 硬件</td><td>启动本地文本模型和最新的简易英语纯文本维基百科存档。视觉模型是可选的,只有点击它的下载按钮时才会下载。模型传输共享一个本地工作者并按顺序运行。</td><td><strong>就绪</strong>要求文本模型和离线资料库完成;视觉模型是可选的。应急箱在它们完成后解锁。</td></tr><tr><td>Firefox</td><td>提供离线维基百科库、阅读器和应急箱。不显示仅限 Chromium 的 WebGPU 文本和视觉下载栈。</td><td>在不假装 Firefox 拥有 Chromium WebGPU 运行时的情况下,存档和文档控件仍然可用。</td></tr></tbody></table></div><p>如果文本和视觉卡片没有显示,请首先确认你使用的是具有 WebGPU 支持的硬件和浏览器版本上的 Chromium 版本。</p></section>- <section class="doc-section" id="setup"><h2>构建基础工具包</h2><ol class="step-list"><li><strong>打开末日模式。</strong>使用设置标题中的核符号末日模式控件。</li><li><strong>检查存储和连接。</strong>Chromium 基础套件包括 1.55 GB 文本模型和当前的简易英语维基百科存档。浏览器内视觉模型是可选的:Ternary Bonsai 2 2B Vision 约 1.45 GB,WebBrain VL 2 450M 约 810 MB。为临时下载状态和应急箱资源留出额外空间。</li><li><strong>开启模式。</strong>在支持的 Chromium 系统上,WebBrain 会自动启动文本模型和所需的维基百科存档,但从不自行下载视觉模型。</li><li><strong>保持浏览器打开以进行模型下载。</strong>你可以切换标签页或关闭设置,但在本地模型传输继续期间必须保持 Chrome 打开。跟踪每个项目的进度,不要仅依赖总存储数字。</li><li><strong>解决任何未就绪的项目。</strong>恢复暂停的项目、重试错误或释放存储空间。移除已完成的模型会删除其缓存文件,需要在离线使用前再次下载。</li><li><strong>测试工具包。</strong>使用视觉测试,打开维基百科阅读器,搜索一篇文章,并打开至少一个已安装的 PDF。下载不等于计划,除非你已证明可以阅读它。</li></ol></section>+ <section class="doc-section" id="setup"><h2>构建基础工具包</h2><ol class="step-list"><li><strong>打开末日模式。</strong>使用设置标题中的核符号末日模式控件。</li><li><strong>检查存储和连接。</strong>Chromium 基础套件包括一个 2.2 GB、也能读取图像的文本模型,以及当前的简易英语维基百科存档。单独的浏览器内视觉模型是可选的:Ternary Bonsai 2 2B Vision 约 1.45 GB,WebBrain VL 2 450M 约 810 MB。为临时下载状态和应急箱资源留出额外空间。</li><li><strong>开启模式。</strong>在支持的 Chromium 系统上,WebBrain 会自动启动文本模型和所需的维基百科存档,但从不自行下载视觉模型。</li><li><strong>保持浏览器打开以进行模型下载。</strong>你可以切换标签页或关闭设置,但在本地模型传输继续期间必须保持 Chrome 打开。跟踪每个项目的进度,不要仅依赖总存储数字。</li><li><strong>解决任何未就绪的项目。</strong>恢复暂停的项目、重试错误或释放存储空间。移除已完成的模型会删除其缓存文件,需要在离线使用前再次下载。</li><li><strong>测试工具包。</strong>使用视觉测试,打开维基百科阅读器,搜索一篇文章,并打开至少一个已安装的 PDF。下载不等于计划,除非你已证明可以阅读它。</li></ol></section>- <section class="doc-section" id="models"><h2>本地模型的功能</h2><div class="card-grid"><div class="card"><span class="tab-key">文本 · 1.55 GB</span><h3>LFM2.5 2.6B</h3><p>WebBrain 下载 <code>LiquidAI/LFM2.5-2.6B-ONNX</code> 的 <code>q4f16</code> 形式,用于通过 WebGPU 进行本地文本生成。就绪后在独立聊天中使用核控件。</p></div><div class="card"><span class="tab-key">视觉 · 可选</span><h3>Ternary Bonsai 2 2B Vision 或 WebBrain VL 2 450M</h3><p>本地视觉层在设备上解读截图,将视觉证据转化为上下文,无需将图像发送到远程视觉提供商。视觉模型卡片与<strong>提供商 → 视觉</strong>共用同一个选项:<strong>自动</strong>(默认)会在 Ternary Bonsai 2 2B Vision(约 1.45 GB)下载完成后使用它;你也可以选择 WebBrain VL 2 450M(约 810 MB,仅用于视觉,需先在提供商页面中同意)或关闭。如果文本模型是 Ternary Bonsai 2 2B Vision,它本身就能读取图像,无需额外下载。</p></div></div><div class="callout warning"><strong class="callout-title">末日模式不会替代你选择的提供商</strong><p>下载的文本模型是独立聊天的运行时选项。启用或下载它不会静默更改普通 WebBrain 对话使用的提供商。</p></div></section>+ <section class="doc-section" id="models"><h2>本地模型的功能</h2><div class="card-grid"><div class="card"><span class="tab-key">文本 · 2.2 GB</span><h3>Compass VLD Tiny v5</h3><p>WebBrain 下载 <code>webbrain-one/webbrain-compass-vld-tiny-v5-onnx</code>(WebBrain 基于 MiniCPM-V-4.7-1B 的微调模型),通过 WebGPU 在本地聊天并调用工具。同一个模型也能读取截图并回答 WebBrain 的决策问题。就绪后在独立聊天中使用核控件。</p></div><div class="card"><span class="tab-key">视觉 · 可选</span><h3>Ternary Bonsai 2 2B Vision 或 WebBrain VL 2 450M</h3><p>本地视觉层在设备上解读截图,将视觉证据转化为上下文,无需将图像发送到远程视觉提供商。视觉模型卡片与<strong>提供商 → 视觉</strong>共用同一个选项:<strong>自动</strong>(默认)会在 Compass VLD Tiny v5 下载完成后使用它,只下载了 Ternary Bonsai 2 2B Vision(约 1.45 GB)时使用后者;你也可以选择 WebBrain VL 2 450M(约 810 MB,仅用于视觉,需先在提供商页面中同意)或关闭。如果文本模型是 Compass VLD Tiny v5 或 Ternary Bonsai 2 2B Vision,它本身就能读取图像,无需额外下载。</p></div></div><div class="callout warning"><strong class="callout-title">末日模式不会替代你选择的提供商</strong><p>下载的文本模型是独立聊天的运行时选项。启用或下载它不会静默更改普通 WebBrain 对话使用的提供商。</p></div></section><section class="doc-section" id="wikipedia"><h2>保留你可以真正搜索的维基百科</h2><p>基础设置下载最新的紧凑型纯文本简易英语存档。当你需要其他语言、更完整版本或包含图像的版本时,打开<strong>管理存档</strong>。WebBrain 会在安装前解析存档的当前版本、日期和确切大小。</p><div class="table-wrap"><table><thead><tr><th>选项</th><th>权衡</th><th>适用场景</th></tr></thead><tbody><tr><td>纯文本</td><td>更小、下载更快;不含完整图像负载的文章文本。</td><td>紧凑参考、本地事实检索和广泛的语言覆盖。</td></tr><tr><td>包含图像</td><td>使用大量存储空间。</td><td>图表、地图和照片很重要的文章。</td></tr><tr><td>导入 <code>.zim</code> 文件</td><td>你管理源文件,重启后可能需要重新授权访问。</td><td>现有的 Kiwix 集合或在机器之间移动的存档。</td></tr></tbody></table></div><p>维基百科阅读器搜索文章标题并从本地存储打开匹配的文章。本地独立聊天可以从已安装的存档中检索紧凑的、有出处的段落。当你替换必需的简易英语存档时,WebBrain 会保留旧副本,直到替换下载并验证成功。</p></section>diff --git a/web/docs/zh/providers/index.html b/web/docs/zh/providers/index.htmlindex 5d8a7dccd..3f57387af 100644--- a/web/docs/zh/providers/index.html+++ b/web/docs/zh/providers/index.html@@ -20,7 +20,7 @@<section class="doc-section" id="local"><h2>本地模型的额外环节</h2><div class="table-wrap"><table><thead><tr><th>服务器</th><th>常见 Base URL</th><th>API 密钥</th></tr></thead><tbody><tr><td>llama.cpp</td><td><code>http://localhost:8080</code></td><td>通常不需要</td></tr><tr><td>Ollama</td><td><code>http://localhost:11434/v1</code></td><td>不需要</td></tr><tr><td>ODS</td><td><code>http://localhost:11434/v1</code></td><td>可选;使用 ODS .env 中的端口</td></tr><tr><td>LM Studio</td><td><code>http://localhost:1234/v1</code></td><td>不需要</td></tr><tr><td>Jan</td><td><code>http://localhost:1337/v1</code></td><td>可选</td></tr><tr><td>vLLM</td><td><code>http://localhost:8000/v1</code></td><td>可选</td></tr><tr><td>SGLang</td><td><code>http://localhost:30000/v1</code></td><td>可选</td></tr></tbody></table></div><div class="callout warning"><strong class="callout-title">终端可用,不代表扩展一定可访问</strong><p>连接被拒绝通常是服务器未启动或端口错误;CORS/Origin 错误需要允许浏览器扩展来源。Ollama 常用:<code>OLLAMA_ORIGINS="chrome-extension://*,moz-extension://*" ollama serve</code>。</p></div></section><section class="doc-section" id="tiers"><h2>提示层级</h2><div class="mode-strip"><div class="mode ask"><strong>Compact</strong><span>最短提示与最小工具面,适合约 8k 上下文的小型本地模型。</span></div><div class="mode act"><strong>Mid</strong><span>包含常见浏览器任务、iframe、下载、计划任务和表单验证。推荐的本地起点。</span></div><div class="mode dev"><strong>Full</strong><span>增加 hover、拖放、框架与 Shadow DOM 等高级回退工具。</span></div></div><p>云提供商固定使用 Full;本地提供商默认 Mid;Compact 提供商不能使用 Dev 模式。</p></section><section class="doc-section" id="troubleshoot"><h2>测试连接失败时</h2><ol><li><strong>Connection refused:</strong>启动本地服务器并确认端口。</li><li><strong>CORS/Origin:</strong>在服务器中允许 WebBrain 扩展来源。</li><li><strong>Unauthorized:</strong>确认密钥与 Base URL 属于同一提供商。</li><li><strong>Model not found:</strong>复制精确模型 ID,或使用 Load models。</li><li><strong>超时:</strong>为大型 CPU 本地模型提高“设置 → 预算”中的 LLM 请求超时。</li><li><strong>几轮后失败:</strong>确认真实上下文窗口,降低提示层级或扩大模型上下文。</li></ol></section>- <section class="doc-section" id="assistive-models"><h2>可选的辅助模型</h2><p>当前提供商负责主要对话、规划和最终回复。提供商页面的其他标签页可独立配置<strong>视觉</strong>模型、<strong>语音</strong>(语音转文字)端点、<strong>生成</strong>媒体、<strong>决策</strong>模型和 <strong>SafeSocial</strong>。这些设置不受提供商筛选或切换影响。</p><p>视觉和决策模型默认都是<strong>自动</strong>:在 Chrome 中,Ternary Bonsai 2 2B Vision 下载完成后使用它(WebBrain Cloud 激活时决策由 Compass 的托管决策模型处理),其他情况下下载前不使用。决策模型只回答“这个任务完成了吗?”这类是非题和选择题;选择<strong>开启</strong>可让 OpenRouter、TypeSafe、本地服务器、D1 或浏览器内模型回答所有决策,选择<strong>关闭</strong>则从不使用。现有权限与完成检查仍然适用。配置方法和数据共享说明见<a href="/docs/zh/settings/#multimodal">视觉、语音与决策模型</a>。</p></section>+ <section class="doc-section" id="assistive-models"><h2>可选的辅助模型</h2><p>当前提供商负责主要对话、规划和最终回复。提供商页面的其他标签页可独立配置<strong>视觉</strong>模型、<strong>语音</strong>(语音转文字)端点、<strong>生成</strong>媒体、<strong>决策</strong>模型和 <strong>SafeSocial</strong>。这些设置不受提供商筛选或切换影响。</p><p>视觉和决策模型默认都是<strong>自动</strong>:在 Chrome 中,Compass VLD Tiny v5 下载完成后使用它(只下载了 Ternary Bonsai 2 2B Vision 或 D1 时,截图用前者、决策用后者;WebBrain Cloud 激活时决策由 Compass 的托管决策模型处理),下载前不使用。决策模型只回答“这个任务完成了吗?”这类是非题和选择题;选择<strong>开启</strong>可让 OpenRouter、TypeSafe、本地服务器、D1 或浏览器内模型回答所有决策,选择<strong>关闭</strong>则从不使用。现有权限与完成检查仍然适用。配置方法和数据共享说明见<a href="/docs/zh/settings/#multimodal">视觉、语音与决策模型</a>。</p></section><nav class="next-links" aria-label="后续指南"><a href="/docs/zh/settings/"><span>上一篇</span>← 设置说明</a><a href="/docs/zh/safety/"><span>下一篇</span>模式、安全与隐私 →</a></nav></main><aside class="on-this-page" aria-label="本页内容"><h2>本页内容</h2><a href="#choice">快速选择</a><a href="#catalog">提供商目录</a><a href="#setup">配置步骤</a><a href="#fields">字段</a><a href="#local">本地模型</a><a href="#tiers">提示层级</a><a href="#troubleshoot">排错</a></aside></div><footer class="site-footer"><div class="footer-inner"><span>WebBrain 用户指南</span><div class="footer-links"><a href="/privacy">隐私</a><a href="https://github.com/webbrain-one/webbrain/issues">报告问题</a></div></div></footer><script src="/docs/assets/docs.js?v=20260809"></script>diff --git a/web/docs/zh/settings/index.html b/web/docs/zh/settings/index.htmlindex 8e6455e30..5c7d695ec 100644--- a/web/docs/zh/settings/index.html+++ b/web/docs/zh/settings/index.html@@ -13,7 +13,7 @@<a class="skip-link" href="#main">跳到正文</a><header class="site-header"><div class="header-inner"><a class="brand" href="/"><img src="/logo-github.png" alt=""><span>WebBrain</span><span class="brand-docs">用户指南</span></a><nav class="top-nav" aria-label="主导航"><a href="/">首页</a><a href="/docs/zh/" aria-current="page">文档</a><a href="/blog/">博客</a><a href="https://github.com/webbrain-one/webbrain">GitHub</a></nav><div class="header-tools"><a class="lang-switch" href="/docs/settings/" aria-label="Read in English"><span>EN</span><span class="active">中文</span></a><button class="icon-button" type="button" data-theme-toggle aria-label="切换主题">☀</button><button class="sidebar-toggle" type="button" data-sidebar-toggle aria-expanded="false" aria-label="打开文档菜单">☰</button></div></div></header><div class="docs-shell"><aside class="docs-sidebar" aria-label="文档目录"><div class="sidebar-group"><div class="sidebar-label">从这里开始</div><nav><a href="/docs/zh/">概览</a><a href="/docs/zh/faq/">常见问题</a></nav></div><div class="sidebar-group"><div class="sidebar-label">使用 WebBrain</div><nav><a href="/docs/zh/settings/" aria-current="page">设置说明</a><a href="/docs/zh/providers/">提供商与模型</a><a href="/docs/zh/easy-cli-proxy/">订阅代理</a><a href="/docs/zh/safety/">模式、安全与隐私</a><a href="/docs/zh/apocalypse-mode/">末日模式</a><a href="/docs/zh/formats/">导出与工作流</a><a href="/docs/zh/mcp/">MCP 服务器</a><a href="/docs/zh/lm-studio/">LM Studio 插件</a><a href="/docs/zh/ollama/">Ollama</a></nav></div><div class="sidebar-group"><div class="sidebar-label">社区</div><nav><a href="https://discord.gg/cgC325ssfw">Discord 支持</a></nav></div></aside><main class="doc-main" id="main"><div class="breadcrumb"><a href="/docs/zh/">文档</a> / 设置</div><section class="doc-hero"><div class="eyebrow">不再猜测设置含义</div><h1>知道该改什么,也知道什么可以不动。</h1><p class="lede">从侧边栏齿轮或浏览器扩展选项打开设置。聊天模型及其辅助模型有单独的<strong>提供商</strong>页面:在侧边栏模型菜单中选择<strong>更多…</strong>,或点击设置页顶部的“提供商”链接即可打开。网站权限也有单独的页面:在麦克风旁边的 <strong>⚠</strong> 菜单中选择“管理网站权限”即可打开。大多数人主要使用“通用”、提供商和权限;其他标签页用于解决特定问题。</p><div class="meta">截图来自设置重新整理后的英文界面 · 中文界面的控件位置相同 · 合成示例不含真实凭据</div></section>- <section class="doc-section" id="defaults"><h2>适合大多数人的默认值</h2><div class="table-wrap"><table><thead><tr><th>控制项</th><th>推荐值</th><th>原因</th></tr></thead><tbody><tr><td>执行前计划</td><td>尝试计划</td><td>有计划时使用;如果规划器格式化失败,安全回退到只读轮次。</td></tr><tr><td>计划审查</td><td>低于置信度阈值时停止</td><td>不确定的计划暂停,可靠计划可继续。</td></tr><tr><td>审查阈值</td><td>90%</td><td>在监督与流畅度之间较保守的平衡。</td></tr><tr><td>最大代理步骤</td><td>130</td><td>足够处理多步骤工作;出现重复时仍应手动停止。</td></tr><tr><td>LLM 请求超时</td><td>120 秒</td><td>云模型通常更快,大型本地模型启动可能更慢。</td></tr><tr><td>自动截图</td><td>导航时</td><td>代理打开页面后截一张图,而不是每次点击或输入后都截图。需要视觉模型检查每个操作时再选“状态变化时”。</td></tr><tr><td>云端费用额度</td><td>每个会话 $10,总计 $10</td><td>付费模型调用的机械停止点,可在“预算”中调高或调低。</td></tr><tr><td>计划任务关键操作确认</td><td>开启</td><td>无人值守任务仍会在关键操作前暂停。</td></tr><tr><td>权限闸门</td><td>开启</td><td>首次在网站执行特定操作前询问,并记住你的明确选择。侧边栏输入框中麦克风旁边的 ⚠ 按钮也可以切换它。</td></tr><tr><td>决策模型</td><td>自动</td><td>无需设置。Ternary Bonsai 2 2B Vision 下载完成后用它做完成检查(WebBrain Cloud 激活时使用 Compass 的托管决策模型),其他情况下不使用决策模型。</td></tr><tr><td>浏览器内视觉模型</td><td>自动</td><td>Ternary Bonsai 2 2B Vision 下载完成后用它读取截图,且仅在聊天模型无法看图时使用。</td></tr></tbody></table></div></section>+ <section class="doc-section" id="defaults"><h2>适合大多数人的默认值</h2><div class="table-wrap"><table><thead><tr><th>控制项</th><th>推荐值</th><th>原因</th></tr></thead><tbody><tr><td>执行前计划</td><td>尝试计划</td><td>有计划时使用;如果规划器格式化失败,安全回退到只读轮次。</td></tr><tr><td>计划审查</td><td>低于置信度阈值时停止</td><td>不确定的计划暂停,可靠计划可继续。</td></tr><tr><td>审查阈值</td><td>90%</td><td>在监督与流畅度之间较保守的平衡。</td></tr><tr><td>最大代理步骤</td><td>130</td><td>足够处理多步骤工作;出现重复时仍应手动停止。</td></tr><tr><td>LLM 请求超时</td><td>120 秒</td><td>云模型通常更快,大型本地模型启动可能更慢。</td></tr><tr><td>自动截图</td><td>导航时</td><td>代理打开页面后截一张图,而不是每次点击或输入后都截图。需要视觉模型检查每个操作时再选“状态变化时”。</td></tr><tr><td>云端费用额度</td><td>每个会话 $10,总计 $10</td><td>付费模型调用的机械停止点,可在“预算”中调高或调低。</td></tr><tr><td>计划任务关键操作确认</td><td>开启</td><td>无人值守任务仍会在关键操作前暂停。</td></tr><tr><td>权限闸门</td><td>开启</td><td>首次在网站执行特定操作前询问,并记住你的明确选择。侧边栏输入框中麦克风旁边的 ⚠ 按钮也可以切换它。</td></tr><tr><td>决策模型</td><td>自动</td><td>无需设置。Compass VLD Tiny v5 下载完成后用它做完成检查,只下载了 D1 时使用 D1(两者都未下载且 WebBrain Cloud 激活时使用 Compass 的托管决策模型),其他情况下不使用决策模型。</td></tr><tr><td>浏览器内视觉模型</td><td>自动</td><td>Compass VLD Tiny v5 下载完成后用它读取截图(只下载了 Ternary Bonsai 2 2B Vision 时用后者),且仅在聊天模型无法看图时使用。</td></tr></tbody></table></div></section><section class="doc-section" id="general"><div class="eyebrow">日常使用</div><h2>通用:代理如何行动</h2><p>外观和下载位置只影响使用体验;下面的控制项会改变代理的工作方式:如何计划、何时由你审查、截取哪些截图。语言和缩放在侧边栏中选择,这里不再重复。</p><figure class="shot"><a href="/docs/assets/screenshots/settings-general.jpg"><img src="/docs/assets/screenshots/settings-general.jpg" alt="WebBrain 通用设置英文界面,包含搜索框、外观、下载文件夹、计划与计划审查"></a><figcaption>“通用”标签页顶部是最常用的控制项。点击可查看原图。</figcaption></figure><p><strong>尝试计划</strong>适合日常使用;<strong>严格计划</strong>会在无法生成有效计划时取消任务;<strong>关闭</strong>会省去一次规划调用,但也失去计划审查节点。当“自动”为某条消息选择 Act 或 Dev,并确定它是简短、直接的任务(例如打开页面、搜索、填写并提交一个表单)时,尝试计划会跳过完整计划,只做快速的意图检查。包含多个目标、大量条目、多个网站、定时执行,或付款、删除等有风险步骤的任务仍会生成计划;计划审查设为始终询问时,每个 Act 或 Dev 轮次也都会生成计划。</p><p><strong>自动截图</strong>默认为<strong>导航时</strong>:代理打开页面后截一张图,让视觉模型看到它到达了哪里。<strong>状态变化时</strong>还会在点击、输入等改变页面的操作后截图,图像令牌消耗更多;<strong>每一步</strong>在每次工具调用后截图;<strong>关闭</strong>只在你要求时截图。</p><p>标签栏下方的搜索框会同时搜索“通用”“预算”“CAPTCHA”和“高级”(无论你当前在其中哪个标签页),并把每个结果显示在所属标签页名称下。显示结果时,不会把任何标签页标记为当前标签页。点击结果上方的标签页名称即可打开该标签页,也可以在标签栏中选择标签页;两种方式都会结束搜索。清空搜索也会结束搜索,并回到你原来所在的标签页。</p></section><section class="doc-section" id="budgets"><div class="eyebrow">限制</div><h2>预算:单个任务能走多远</h2><p>“预算”把所有关于花费、等待和工作量的限制集中在一个标签页:云端费用额度、问题等待你回答的时间、步骤上限、模型请求超时和截图的图像预算。</p><figure class="shot"><a href="/docs/assets/screenshots/settings-budgets.jpg"><img src="/docs/assets/screenshots/settings-budgets.jpg" alt="WebBrain 预算设置英文界面,包含云端费用额度、问题超时、步骤上限、请求超时和图像预算"></a><figcaption>所有机械停止点都在这里。示例值为默认值。</figcaption></figure><div class="table-wrap"><table><thead><tr><th>设置</th><th>默认值</th><th>限制什么</th></tr></thead><tbody><tr><td>云端费用会话额度</td><td>$10.00</td><td>单个对话中云端和路由模型调用的估算花费。达到额度后,WebBrain 会在下一次调用前停止。</td></tr><tr><td>云端费用总额度</td><td>$10.00</td><td>所有对话的估算总花费。标签会显示已花费金额,例如 <code>$0.00 / $10.00</code>;<strong>重置花费</strong>会重新开始计数。</td></tr><tr><td>澄清超时</td><td>60 秒</td><td>问题卡片等待你回答的时间,超时后 WebBrain 自行作答。滑块两端分别是“立即”和“关闭”。</td></tr><tr><td>问题超时时</td><td>完成任务</td><td>超时后的答案是继续推进任务,还是选择更安全的做法。</td></tr><tr><td>代理最大步骤数</td><td>130</td><td>一条消息中的工具调用步骤,不是聊天消息数。</td></tr><tr><td>LLM 请求超时</td><td>120 秒</td><td>等待模型开始响应的时间。</td></tr><tr><td>图像预算</td><td>自动细节、每轮不限、1568 px</td><td>每轮最多向视觉模型发送多少张截图,以及截图尺寸。</td></tr></tbody></table></div><p>步骤上限越高,长流程越有机会完成和恢复,但也可能消耗更多令牌。请求超时只限制模型开始响应前的等待时间,不会中断已经开始流式输出的响应。</p><div class="callout warning"><strong class="callout-title">无限步骤不等于更可靠</strong><p>∞ 只会移除机械停止点,不会让模型更不容易循环。除非你正在全程监督长任务,否则保留有限上限。</p></div></section><section class="doc-section" id="memory"><div class="eyebrow">按需使用</div><h2>记忆:决定哪些信息长期保留</h2><figure class="shot"><a href="/docs/assets/screenshots/settings-memory.jpg"><img src="/docs/assets/screenshots/settings-memory.jpg" alt="WebBrain 记忆设置英文界面,显示加密云同步与个人资料自动填充"></a><figcaption>记忆标签页包含加密同步、个人资料自动填充和用户记忆三种不同功能。</figcaption></figure><ul><li><strong>加密云同步</strong>面向 WebBrain Compass 会员;支持的提供商密钥、个人资料和记忆会在上传前用同步密码加密。聊天历史与 OAuth 登录不会同步。</li><li><strong>个人资料自动填充</strong>以明文保存在当前浏览器配置中,启用后会发送给当前模型提供商。</li><li><strong>用户记忆</strong>保存稳定偏好;<code>/memory --add</code> 会立即保存,自动学习是可选项。</li></ul><div class="callout danger"><strong class="callout-title">不要把记忆当作密码管理器</strong><p>不要在个人资料或用户记忆中存放密码、API 密钥、令牌、恢复码等秘密。</p></div></section>@@ -21,11 +21,11 @@<section class="doc-section" id="captcha"><div class="eyebrow">按需使用</div><h2>CAPTCHA:求解服务账户</h2><p>CAPTCHA 标签页为 CapSolver、2Captcha、CapMonster Cloud、SolveCaptcha、Anti-Captcha、NopeCHA 和 NoneCap 各提供一张卡片。输入密钥会勾选启用;你也可以取消勾选并保留密钥。保存后生效,测试余额不会自动启用服务。</p><figure class="shot"><a href="/docs/assets/screenshots/settings-captcha.jpg"><img src="/docs/assets/screenshots/settings-captcha.jpg" alt="WebBrain CAPTCHA 设置英文界面,每个求解服务一张卡片"></a><figcaption>保存密钥前不会启用任何服务;测试余额也不会启用求解。</figcaption></figure><p>展开每个服务的高级设置可修改权重,默认依次为 100 至 94。可同时启用多个服务,兼容的服务按权重从高到低尝试。错误或超时后的回退可能产生多次费用。hCaptcha 使用 NopeCHA 或 NoneCap;其他类型需要相应服务支持和页面参数。请阅读<a href="/docs/captcha/">完整 CAPTCHA 指南(英文)</a>。</p></section><section class="doc-section" id="advanced"><div class="eyebrow">专家标签页</div><h2>高级:带着明确目的再打开</h2><p>“高级”现在是一个独立的标签页。用搜索框可以直接跳到某个控制项;它会同时搜索“高级”“通用”“预算”和“CAPTCHA”。</p><figure class="shot"><a href="/docs/assets/screenshots/settings-advanced.jpg"><img src="/docs/assets/screenshots/settings-advanced.jpg" alt="WebBrain 高级设置英文界面,包含帮助改进 WebBrain、记录轨迹和本地反馈诊断"></a><figcaption>每个控制项旁边都保留完整说明。</figcaption></figure><div class="table-wrap"><table><thead><tr><th>设置</th><th>通常保持</th><th>何时修改</th></tr></thead><tbody><tr><td>记录轨迹</td><td>关闭</td><td>比较模型或诊断运行,并接受本地存储。一个开关会连同完整请求细节(提示、消息、工具定义和结果)一起记录每次运行。</td></tr><tr><td>截图回退</td><td>视觉模型可用时开启</td><td>DOM 读取遗漏图表、画布或视觉内容。</td></tr><tr><td>截图打码</td><td>关闭</td><td>希望在截图发送给视觉模型前,先模糊表单字段和看起来像邮箱或电话号码的文字。这是尽力而为的保护,不是保证。</td></tr><tr><td>站点适配器</td><td>开启</td><td>调试令牌使用或禁用站点专用提示。</td></tr><tr><td>API 变更观察器</td><td>关闭</td><td>研究重复 UI 操作与 API 快捷建议。</td></tr><tr><td>严格的机密处理</td><td>个人日常使用时关闭</td><td>屏幕共享或导出轨迹,不希望凭据被引用。</td></tr><tr><td>允许代理访问本地网络</td><td>关闭</td><td>必须访问 localhost、内网或私有地址。</td></tr></tbody></table></div></section><section class="doc-section" id="providers"><div class="eyebrow">日常使用</div><h2>提供商:选择当前模型</h2><p>提供商是一个独立的页面。在侧边栏模型菜单中选择<strong>更多…</strong>,或点击设置页顶部的<strong>提供商</strong>链接即可打开。第一个标签页 <strong>LLM</strong> 列出聊天提供商,其他标签页是下面介绍的辅助模型。</p><p>一次只有一个提供商处于活动状态。正确顺序是 <strong>保存 → 测试连接 → 设为活动</strong>。</p><figure class="shot"><a href="/docs/assets/screenshots/settings-providers.jpg"><img src="/docs/assets/screenshots/settings-providers.jpg" alt="WebBrain 提供商页面 LLM 标签页英文界面,包含分类筛选、搜索和当前活动提供商"></a><figcaption>即使使用分类筛选,当前活动提供商也会保留在列表中。</figcaption></figure><p>本地服务器地址、上下文窗口和提示层级见<a href="/docs/zh/providers/">提供商与模型</a>。</p></section>- <section class="doc-section" id="multimodal"><div class="eyebrow">按需使用</div><h2>视觉、语音与决策模型:支持主要代理</h2><p>在提供商页面的 <strong>LLM</strong> 标签页中选择主要对话模型。该页面的其他标签页是可选的辅助模型:<strong>视觉</strong>、<strong>语音</strong>、<strong>生成</strong>(图像和视频)、<strong>决策</strong>和 <strong>SafeSocial</strong>。配置它们不会改变当前提供商。</p><p>视觉和决策模型默认都是<strong>自动</strong>。在支持硬件 WebGPU 的 Chrome 中,自动模式会在 <strong>Ternary Bonsai 2 2B Vision</strong>(约 1.45 GB,在本机运行)下载完成后使用它(WebBrain Cloud 激活时改由 Compass 的托管决策模型回答),其他情况下下载前不使用。安装页上的<strong>下载本地模型</strong>按钮、<strong>提供商 → WebGPU</strong>、视觉卡片和末日模式下载的都是同一个模型,一次下载即可同时用于聊天、截图和决策。</p>- <h3>视觉</h3><p>选择由谁读取截图。保存了 OpenAI 兼容端点时,所有截图都交给它,当前提供商继续负责规划和工具调用。没有端点时,当前模型若能看图就自己读取截图;不能看图时,由浏览器内视觉模型读取(仅 Chrome)。全部保持默认即可让当前提供商处理视觉。截图限制在“设置 → 预算”中,截图打码在“设置 → 高级”中。</p><p>浏览器内视觉模型有四个选项:</p><ul><li><strong>自动</strong>(默认):Ternary Bonsai 2 2B Vision 下载完成后使用它,下载前保持关闭。</li><li><strong>Ternary Bonsai 2 2B Vision</strong>:约 1.45 GB,同一个模型也能聊天和回答决策问题。</li><li><strong>WebBrain VL 2 450M</strong>:约 810 MB,只用于视觉。首次下载会征求你的同意并检查 WebGPU。</li><li><strong>关闭</strong>:不使用浏览器内视觉模型。</li></ul><p>更改选项不会自动下载任何内容,请使用卡片上的<strong>下载</strong>按钮。任务从不等待下载完成;在浏览器内读取的截图不会离开你的设备。Firefox 没有浏览器内视觉模型。</p><figure class="shot"><a href="/docs/assets/screenshots/providers-vision.jpg"><img src="/docs/assets/screenshots/providers-vision.jpg" alt="提供商页面视觉标签页英文界面:浏览器内视觉模型设为自动,下方是可选的 OpenAI 兼容端点"></a><figcaption>全新安装:浏览器内视觉模型为自动,在 Ternary Bonsai 2 2B Vision 下载完成前保持关闭。</figcaption></figure>+ <section class="doc-section" id="multimodal"><div class="eyebrow">按需使用</div><h2>视觉、语音与决策模型:支持主要代理</h2><p>在提供商页面的 <strong>LLM</strong> 标签页中选择主要对话模型。该页面的其他标签页是可选的辅助模型:<strong>视觉</strong>、<strong>语音</strong>、<strong>生成</strong>(图像和视频)、<strong>决策</strong>和 <strong>SafeSocial</strong>。配置它们不会改变当前提供商。</p><p>视觉和决策模型默认都是<strong>自动</strong>。在支持硬件 WebGPU 的 Chrome 中,自动模式会在 <strong>Compass VLD Tiny v5</strong>(约 2.2 GB,在本机运行)下载完成后使用它,下载前不使用;WebBrain Cloud 激活时,在它下载完成前由 Compass 的托管决策模型回答决策。安装页上的<strong>下载本地模型</strong>按钮、<strong>提供商 → WebGPU</strong>、视觉卡片和末日模式下载的都是同一个模型,一次下载即可同时用于聊天、截图和决策。</p>+ <h3>视觉</h3><p>选择由谁读取截图。保存了 OpenAI 兼容端点时,所有截图都交给它,当前提供商继续负责规划和工具调用。没有端点时,当前模型若能看图就自己读取截图;不能看图时,由浏览器内视觉模型读取(仅 Chrome)。全部保持默认即可让当前提供商处理视觉。截图限制在“设置 → 预算”中,截图打码在“设置 → 高级”中。</p><p>浏览器内视觉模型有五个选项:</p><ul><li><strong>自动</strong>(默认):Compass VLD Tiny v5 下载完成后使用它,只下载了 Ternary Bonsai 2 2B Vision 时使用后者,下载前保持关闭。</li><li><strong>Compass VLD Tiny v5</strong>:约 2.2 GB,同一个模型也能聊天和回答决策问题。</li><li><strong>Ternary Bonsai 2 2B Vision</strong>:约 1.45 GB,同一个模型也能聊天。</li><li><strong>WebBrain VL 2 450M</strong>:约 810 MB,只用于视觉。首次下载会征求你的同意并检查 WebGPU。</li><li><strong>关闭</strong>:不使用浏览器内视觉模型。</li></ul><p>更改选项不会自动下载任何内容,请使用卡片上的<strong>下载</strong>按钮。任务从不等待下载完成;在浏览器内读取的截图不会离开你的设备。Firefox 没有浏览器内视觉模型。</p><figure class="shot"><a href="/docs/assets/screenshots/providers-vision.jpg"><img src="/docs/assets/screenshots/providers-vision.jpg" alt="提供商页面视觉标签页英文界面:浏览器内视觉模型设为自动,下方是可选的 OpenAI 兼容端点"></a><figcaption>全新安装:浏览器内视觉模型为自动,在 Compass VLD Tiny v5 下载完成前保持关闭。</figcaption></figure><h3>语音</h3><p>为录音配置 Whisper 兼容端点。留空时依次尝试已配置的 OpenAI、Groq、LM Studio、llama.cpp。</p><h3>生成</h3><p>为图像和视频生成工具连接 fal.ai、OpenRouter、ComfyRouter 或本地 ComfyUI 服务器。默认不配置任何服务。</p>- <h3>决策</h3><p>决策模型是一个小而快的模型,负责回答智能体的是非题和选择题,例如“这个任务完成了吗?”,这样聊天模型就不必处理这些问题。它从不规划、点击或输入。可选三种模式:</p><ul><li><strong>自动</strong>(默认):无需设置。在 Chrome 中,Ternary Bonsai 2 2B Vision 下载完成后,用它检查侧边栏任务以及计划监控和计划任务是否真正完成。当 WebBrain Cloud(Compass)是你的聊天提供商时,改由 Compass 的决策模型回答。其他情况下不使用决策模型。卡片会显示当前使用的是哪个。</li><li><strong>开启</strong>:所有决策都使用你配置的提供商:OpenRouter、TypeSafe、本地 System One 服务器、D1,或 Chrome 中的浏览器内聊天模型。所有决策默认开启。展开<strong>选择哪些决策使用它</strong>可以单独关闭某些决策(验证完成、验证计划监控、验证计划任务完成、快速分类、快速浏览器操作(实验性))并设置阈值。</li><li><strong>关闭</strong>:从不使用决策模型,包括 Compass 的决策模型。聊天模型仍会检查任务是否完成。</li></ul><p>切换模式立即生效;“开启”模式下的提供商字段需要点击<strong>保存</strong>。<strong>测试连接</strong>只用固定示例数据发送一次请求。<strong>重置</strong>会回到自动模式并删除已保存的决策密钥。云端决策提供商会收到有限的任务与页面数据,可能产生费用;严格秘密模式会禁用决策模型。在这些模式推出前已开启决策外包的配置会按“开启”继续工作,并保留原有开关。</p><figure class="shot"><a href="/docs/assets/screenshots/settings-decision-models.jpg"><img src="/docs/assets/screenshots/settings-decision-models.jpg" alt="决策模型英文设置界面,选中自动,状态行显示尚未使用决策模型"></a><figcaption>全新安装时的自动模式:尚未下载任何模型,因此没有使用决策模型(除非当前提供商是 WebBrain Cloud,此时由 Compass 的决策模型回答)。只有选择“开启”时才会显示提供商字段。</figcaption></figure>+ <h3>决策</h3><p>决策模型是一个小而快的模型,负责回答智能体的是非题和选择题,例如“这个任务完成了吗?”,这样聊天模型就不必处理这些问题。它从不规划、点击或输入。可选三种模式:</p><ul><li><strong>自动</strong>(默认):无需设置。在 Chrome 中,Compass VLD Tiny v5 下载完成后,用它检查侧边栏任务以及计划监控和计划任务是否真正完成;只下载了 D1 时由 D1 检查。当 WebBrain Cloud(Compass)是你的聊天提供商时,在其中一个下载完成前由 Compass 的决策模型回答。其他情况下不使用决策模型。卡片会显示当前使用的是哪个。</li><li><strong>开启</strong>:所有决策都使用你配置的提供商:OpenRouter、TypeSafe、本地 System One 服务器、D1,或 Chrome 中的浏览器内聊天模型。所有决策默认开启。展开<strong>选择哪些决策使用它</strong>可以单独关闭某些决策(验证完成、验证计划监控、验证计划任务完成、快速分类、快速浏览器操作(实验性))并设置阈值。</li><li><strong>关闭</strong>:从不使用决策模型,包括 Compass 的决策模型。聊天模型仍会检查任务是否完成。</li></ul><p>切换模式立即生效;“开启”模式下的提供商字段需要点击<strong>保存</strong>。<strong>测试连接</strong>只用固定示例数据发送一次请求。<strong>重置</strong>会回到自动模式并删除已保存的决策密钥。云端决策提供商会收到有限的任务与页面数据,可能产生费用;严格秘密模式会禁用决策模型。在这些模式推出前已开启决策外包的配置会按“开启”继续工作,并保留原有开关。</p><figure class="shot"><a href="/docs/assets/screenshots/settings-decision-models.jpg"><img src="/docs/assets/screenshots/settings-decision-models.jpg" alt="决策模型英文设置界面,选中自动,状态行显示尚未使用决策模型"></a><figcaption>全新安装时的自动模式:尚未下载任何模型,因此没有使用决策模型(除非当前提供商是 WebBrain Cloud,此时由 Compass 的决策模型回答)。只有选择“开启”时才会显示提供商字段。</figcaption></figure><h3>SafeSocial</h3><p>可选的本机 Instagram 图片分类器。在你开启前保持关闭;模型只下载一次,并在浏览器内运行。</p></section><section class="doc-section" id="permissions"><div class="eyebrow">日常安全</div><h2>权限:WebBrain 可以在哪里做什么</h2><p>权限是一个独立的页面。保持“执行重要操作前询问”开启。你在侧边栏选择“始终允许”时,WebBrain 会记录一个具体网站的具体能力,并在权限页面中列出。每条永久授权都对应一个明确的“网站 + 能力”组合;允许点击并不自动允许输入、下载、上传、执行 JavaScript 或安排任务。</p><p>侧边栏里也有同一个开关:点击输入框中麦克风旁边的 <strong>⚠</strong> 按钮即可打开,里面还有前往权限页面的链接。提示关闭期间,该按钮显示为橙色并带一个圆点。在这里关闭提示不会批准已经在等待的权限卡片,那张卡片仍需你自己回答。</p><figure class="shot"><a href="/docs/assets/screenshots/settings-permissions.jpg"><img src="/docs/assets/screenshots/settings-permissions.jpg" alt="WebBrain 权限页面英文界面,显示总开关与按网站记录的能力授权"></a><figcaption>撤销单项后,WebBrain 下次需要该能力时会重新询问。</figcaption></figure></section>