[{"data":1,"prerenderedAt":39},["ShallowReactive",2],{"doc-en-ai":3},{"meta":4,"html":10,"headings":11,"raw":38},{"title":5,"description":6,"icon":7,"order":8,"updated":9},"AI Explanations","Ask AI what an unfamiliar cleanup item, startup item or system setting does and what changing it would affect, using a model service you connect yourself.","ai",13,"2026-10-29","\u003Cp>Not sure whether a cache can be cleared or a startup item can be turned off? \u003Cstrong>AI explanation\u003C\u002Fstrong> tells you what an item is and what an action could affect. It only explains; it never deletes files or changes settings for you.\u003C\u002Fp>\n\u003Ch2 id=\"where\">\u003Ca class=\"heading-anchor\" href=\"#where\" aria-hidden=\"true\" tabindex=\"-1\">#\u003C\u002Fa>Where to use it\u003C\u002Fh2>\n\u003Cp>AI explanations are available in \u003Cstrong>Deep Cleanup\u003C\u002Fstrong>, \u003Cstrong>Privacy Cleanup\u003C\u002Fstrong>, \u003Cstrong>Large File Cleanup\u003C\u002Fstrong>, \u003Cstrong>Duplicate Cleanup\u003C\u002Fstrong>, \u003Cstrong>App Uninstaller\u003C\u002Fstrong>, \u003Cstrong>Startup Items\u003C\u002Fstrong>, \u003Cstrong>System Optimization\u003C\u002Fstrong> and \u003Cstrong>System Maintenance\u003C\u002Fstrong>. Hover over an item and click \u003Cstrong>AI explanation\u003C\u002Fstrong>; the answer appears in a panel as it is generated. You can \u003Cstrong>Stop generating\u003C\u002Fstrong>, \u003Cstrong>Minimize\u003C\u002Fstrong> the panel to read later, or copy the answer. Custom rules in Deep Cleanup are not supported yet.\u003C\u002Fp>\n\u003Cp>When an explanation finishes, you can \u003Cstrong>ask a follow-up\u003C\u002Fstrong> at the bottom of the panel: pick a suggested question such as "What is it actually used for?" or "What happens if I delete it?", or type your own and press Return to send (Shift+Return adds a line). You can ask up to eight follow-up questions per item, and reopening the same item keeps the conversation.\u003C\u002Fp>\n\u003Ch2 id=\"custom-service\">\u003Ca class=\"heading-anchor\" href=\"#custom-service\" aria-hidden=\"true\" tabindex=\"-1\">#\u003C\u002Fa>Set up a custom service\u003C\u002Fh2>\n\u003Cp>This version does not include a built-in free AI service, and the settings show \u003Cstrong>This version supports custom services only\u003C\u002Fstrong>. Connect your own model service instead:\u003C\u002Fp>\n\u003Col>\n\u003Cli>Open \u003Cstrong>Settings → Scanning & analysis → AI explanations\u003C\u002Fstrong> and click \u003Cstrong>Configure…\u003C\u002Fstrong>, or click \u003Cstrong>Configure\u003C\u002Fstrong> in the AI explanation panel.\u003C\u002Fli>\n\u003Cli>Choose \u003Cstrong>Custom service\u003C\u002Fstrong> and fill in three fields:\u003Cul>\n\u003Cli>\u003Cstrong>Base URL\u003C\u002Fstrong>: your provider's API root, without the trailing \u003Ccode>\u002Fchat\u002Fcompletions\u003C\u002Fcode>. For example, if the full endpoint is \u003Ccode>https:\u002F\u002Fapi.example.com\u002Fv1\u002Fchat\u002Fcompletions\u003C\u002Fcode>, enter \u003Ccode>https:\u002F\u002Fapi.example.com\u002Fv1\u003C\u002Fcode>;\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Model name\u003C\u002Fstrong>: the model ID from your provider, which may differ from the name shown on its website;\u003C\u002Fli>\n\u003Cli>\u003Cstrong>API Key\u003C\u002Fstrong>: a key created in your provider's console.\u003C\u002Fli>\n\u003C\u002Ful>\n\u003C\u002Fli>\n\u003Cli>Click \u003Cstrong>Save and test\u003C\u002Fstrong> to check the connection. The test sends one short request and may incur a small charge. To store the settings without testing, click \u003Cstrong>Save\u003C\u002Fstrong>.\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Cp>Any service with an OpenAI-compatible Chat Completions API that supports streaming works, including compatible endpoints from model providers and API gateways. Pricing and usage limits are set by your provider.\u003C\u002Fp>\n\u003Ch3 id=\"advanced-settings\">\u003Ca class=\"heading-anchor\" href=\"#advanced-settings\" aria-hidden=\"true\" tabindex=\"-1\">#\u003C\u002Fa>Advanced settings\u003C\u002Fh3>\n\u003Cp>The defaults usually work. Set \u003Cstrong>Reasoning mode\u003C\u002Fstrong> to \u003Cstrong>Disable reasoning (compatible providers)\u003C\u002Fstrong> to reduce waiting, and switch back to \u003Cstrong>Provider default\u003C\u002Fstrong> if your provider returns an error. Leave \u003Cstrong>Temperature\u003C\u002Fstrong> and \u003Cstrong>Max Tokens\u003C\u002Fstrong> empty to use the provider's defaults; a low Max Tokens value can cut answers short. If your provider requires extra headers, add up to 8 under \u003Cstrong>Custom request headers\u003C\u002Fstrong>.\u003C\u002Fp>\n\u003Ch2 id=\"local-models\">\u003Ca class=\"heading-anchor\" href=\"#local-models\" aria-hidden=\"true\" tabindex=\"-1\">#\u003C\u002Fa>Use a local model\u003C\u002Fh2>\n\u003Cp>A model service running on your Mac works if it offers an OpenAI-compatible API, with a Base URL such as \u003Ccode>http:\u002F\u002F127.0.0.1:PORT\u002Fv1\u003C\u002Fcode>. For \u003Ccode>localhost\u003C\u002Fcode> or \u003Ccode>127.0.0.1\u003C\u002Fcode> you can leave API Key empty. A service on another computer on your network needs an API Key; if it doesn't check keys, any value works. HTTP connections are not encrypted, so use HTTPS for remote services.\u003C\u002Fp>\n\u003Ch2 id=\"data-sent\">\u003Ca class=\"heading-anchor\" href=\"#data-sent\" aria-hidden=\"true\" tabindex=\"-1\">#\u003C\u002Fa>What is sent\u003C\u002Fh2>\n\u003Cp>Only when you click \u003Cstrong>AI explanation\u003C\u002Fstrong> does WashBear send information about that item, directly from your Mac to the service you configured, without passing through our servers:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>the item's name, description, type and current status, and your interface language;\u003C\u002Fli>\n\u003Cli>relevant full paths, such as file locations, startup configurations and program paths, which may include your macOS user name;\u003C\u002Fli>\n\u003Cli>extra details depending on the feature, such as file size and modification time, a startup item's publisher, version and signature status, an app's identifier, signing details, copyright notice, the reasons it gives when asking for permissions and the product description built into it, or the apps that must close before cleanup;\u003C\u002Fli>\n\u003Cli>for follow-up questions, the earlier answers and questions about that item.\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>File contents are never sent. Requests also carry your API Key, any custom headers, and basic details such as the app and macOS versions. How the provider handles this data is governed by its own terms and privacy policy. Your API Key is stored only on this Mac; click \u003Cstrong>Delete configuration\u003C\u002Fstrong> to remove it.\u003C\u002Fp>\n\u003Ch2 id=\"explain-only\">\u003Ca class=\"heading-anchor\" href=\"#explain-only\" aria-hidden=\"true\" tabindex=\"-1\">#\u003C\u002Fa>AI explains, you decide\u003C\u002Fh2>\n\u003Cp>AI cannot delete files, change settings or run commands; every cleanup and change still needs your confirmation. Answers can be wrong, which is why the panel notes \u003Cstrong>AI is for reference only\u003C\u002Fstrong>. Before handling important data, check its location and make a backup.\u003C\u002Fp>\n\u003Ch2 id=\"turn-off\">\u003Ca class=\"heading-anchor\" href=\"#turn-off\" aria-hidden=\"true\" tabindex=\"-1\">#\u003C\u002Fa>Turn AI off\u003C\u002Fh2>\n\u003Cp>Turn off the switch in \u003Cstrong>Settings → Scanning & analysis → AI explanations\u003C\u002Fstrong> to hide AI explanations on every page. Your saved configuration is kept, so you can turn it back on at any time.\u003C\u002Fp>\n\u003Ch2 id=\"troubleshooting\">\u003Ca class=\"heading-anchor\" href=\"#troubleshooting\" aria-hidden=\"true\" tabindex=\"-1\">#\u003C\u002Fa>Troubleshooting\u003C\u002Fh2>\n\u003Cul>\n\u003Cli>\u003Cstrong>Authentication failed\u003C\u002Fstrong>: check that the API Key is complete and valid, and comes from the same provider as the Base URL.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Model or endpoint not found\u003C\u002Fstrong>: check the Base URL and model name.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Insufficient balance or rate limit reached\u003C\u002Fstrong>: check your balance and limits in the provider's console.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>The provider rejected the request\u003C\u002Fstrong>: set Reasoning mode back to \u003Cstrong>Provider default\u003C\u002Fstrong>.\u003C\u002Fli>\n\u003Cli>Answers cut short: raise or clear Max Tokens. After you stop generating, the provider may still charge for the request.\u003C\u002Fli>\n\u003C\u002Ful>\n",[12,16,19,23,26,29,32,35],{"depth":13,"id":14,"text":15},2,"where","Where to use it",{"depth":13,"id":17,"text":18},"custom-service","Set up a custom service",{"depth":20,"id":21,"text":22},3,"advanced-settings","Advanced settings",{"depth":13,"id":24,"text":25},"local-models","Use a local model",{"depth":13,"id":27,"text":28},"data-sent","What is sent",{"depth":13,"id":30,"text":31},"explain-only","AI explains, you decide",{"depth":13,"id":33,"text":34},"turn-off","Turn AI off",{"depth":13,"id":36,"text":37},"troubleshooting","Troubleshooting","\nNot sure whether a cache can be cleared or a startup item can be turned off? **AI explanation** tells you what an item is and what an action could affect. It only explains; it never deletes files or changes settings for you.\n\n## Where to use it {#where}\n\nAI explanations are available in **Deep Cleanup**, **Privacy Cleanup**, **Large File Cleanup**, **Duplicate Cleanup**, **App Uninstaller**, **Startup Items**, **System Optimization** and **System Maintenance**. Hover over an item and click **AI explanation**; the answer appears in a panel as it is generated. You can **Stop generating**, **Minimize** the panel to read later, or copy the answer. Custom rules in Deep Cleanup are not supported yet.\n\nWhen an explanation finishes, you can **ask a follow-up** at the bottom of the panel: pick a suggested question such as \"What is it actually used for?\" or \"What happens if I delete it?\", or type your own and press Return to send (Shift+Return adds a line). You can ask up to eight follow-up questions per item, and reopening the same item keeps the conversation.\n\n## Set up a custom service {#custom-service}\n\nThis version does not include a built-in free AI service, and the settings show **This version supports custom services only**. Connect your own model service instead:\n\n1. Open **Settings → Scanning & analysis → AI explanations** and click **Configure…**, or click **Configure** in the AI explanation panel.\n2. Choose **Custom service** and fill in three fields:\n   - **Base URL**: your provider's API root, without the trailing `\u002Fchat\u002Fcompletions`. For example, if the full endpoint is `https:\u002F\u002Fapi.example.com\u002Fv1\u002Fchat\u002Fcompletions`, enter `https:\u002F\u002Fapi.example.com\u002Fv1`;\n   - **Model name**: the model ID from your provider, which may differ from the name shown on its website;\n   - **API Key**: a key created in your provider's console.\n3. Click **Save and test** to check the connection. The test sends one short request and may incur a small charge. To store the settings without testing, click **Save**.\n\nAny service with an OpenAI-compatible Chat Completions API that supports streaming works, including compatible endpoints from model providers and API gateways. Pricing and usage limits are set by your provider.\n\n### Advanced settings {#advanced-settings}\n\nThe defaults usually work. Set **Reasoning mode** to **Disable reasoning (compatible providers)** to reduce waiting, and switch back to **Provider default** if your provider returns an error. Leave **Temperature** and **Max Tokens** empty to use the provider's defaults; a low Max Tokens value can cut answers short. If your provider requires extra headers, add up to 8 under **Custom request headers**.\n\n## Use a local model {#local-models}\n\nA model service running on your Mac works if it offers an OpenAI-compatible API, with a Base URL such as `http:\u002F\u002F127.0.0.1:PORT\u002Fv1`. For `localhost` or `127.0.0.1` you can leave API Key empty. A service on another computer on your network needs an API Key; if it doesn't check keys, any value works. HTTP connections are not encrypted, so use HTTPS for remote services.\n\n## What is sent {#data-sent}\n\nOnly when you click **AI explanation** does WashBear send information about that item, directly from your Mac to the service you configured, without passing through our servers:\n\n- the item's name, description, type and current status, and your interface language;\n- relevant full paths, such as file locations, startup configurations and program paths, which may include your macOS user name;\n- extra details depending on the feature, such as file size and modification time, a startup item's publisher, version and signature status, an app's identifier, signing details, copyright notice, the reasons it gives when asking for permissions and the product description built into it, or the apps that must close before cleanup;\n- for follow-up questions, the earlier answers and questions about that item.\n\nFile contents are never sent. Requests also carry your API Key, any custom headers, and basic details such as the app and macOS versions. How the provider handles this data is governed by its own terms and privacy policy. Your API Key is stored only on this Mac; click **Delete configuration** to remove it.\n\n## AI explains, you decide {#explain-only}\n\nAI cannot delete files, change settings or run commands; every cleanup and change still needs your confirmation. Answers can be wrong, which is why the panel notes **AI is for reference only**. Before handling important data, check its location and make a backup.\n\n## Turn AI off {#turn-off}\n\nTurn off the switch in **Settings → Scanning & analysis → AI explanations** to hide AI explanations on every page. Your saved configuration is kept, so you can turn it back on at any time.\n\n## Troubleshooting {#troubleshooting}\n\n- **Authentication failed**: check that the API Key is complete and valid, and comes from the same provider as the Base URL.\n- **Model or endpoint not found**: check the Base URL and model name.\n- **Insufficient balance or rate limit reached**: check your balance and limits in the provider's console.\n- **The provider rejected the request**: set Reasoning mode back to **Provider default**.\n- Answers cut short: raise or clear Max Tokens. After you stop generating, the provider may still charge for the request.\n",1793257449910]