Summarize: dedicated LLM connection profile for summarization Reviewed-by: auto-review

0320ffda52ab4b428632179444577f043defcc1f

permissionBRICK <permissionBRICK@gmail.com>

Parents: acdce0422654c044416be6797fdefd3b99a630ec

public/scripts/extensions/memory/index.js+117 -1 Viewed

 import { SlashCommand } from '../../slash-commands/SlashCommand.js';
 import { ARGUMENT_TYPE, SlashCommandArgument, SlashCommandNamedArgument } from '../../slash-commands/SlashCommandArgument.js';
 import { macros, MacroCategory } from '../../macros/macro-system.js';
-import { countWebLlmTokens, generateWebLlmChatPrompt, getWebLlmContextSize, isWebLlmSupported } from '../shared.js';
+import { ConnectionManagerRequestService, countWebLlmTokens, generateWebLlmChatPrompt, getWebLlmContextSize, isWebLlmSupported } from '../shared.js';
 import { commonEnumProviders } from '../../slash-commands/SlashCommandCommonEnumsProvider.js';
 import { removeReasoningFromString } from '../../reasoning.js';
 import { MacrosParser } from '/scripts/macros.js';

     maxMessagesPerRequestStep: 1,
     prompt_builder: prompt_builders.DEFAULT,
     summaryPromptRole: extension_prompt_roles.SYSTEM,
+    summaryConnectionProfile: '',
 };
 /**

         : extension_prompt_roles.SYSTEM;
+}
+let summaryConnectionProfileDropdownInitialized = false;
+/**
+ * Populates the dedicated summarization connection profile dropdown.
+ * Only initializes once to avoid attaching duplicate Connection Manager event listeners
+ * when loadSettings() is called again (e.g. after loading a settings preset).
+ */
+function initSummaryConnectionProfileDropdown() {
+    if (summaryConnectionProfileDropdownInitialized) {
+        return;
+    }
+    try {
+        ConnectionManagerRequestService.handleDropdown(
+            '#memory_summary_connection_profile',
+            extension_settings.memory.summaryConnectionProfile,
+            (profile) => {
+                extension_settings.memory.summaryConnectionProfile = profile?.id ?? '';
+                saveSettingsDebounced();
+            },
+        );
+        summaryConnectionProfileDropdownInitialized = true;
+    } catch (error) {
+        // Connection Manager may be unavailable/disabled; leave the dropdown empty in that case.
+        console.warn('Summarize: could not populate summary connection profile dropdown', error);
+    }
+}
 function loadSettings() {
     if (Object.keys(extension_settings.memory).length === 0) {
         Object.assign(extension_settings.memory, defaultSettings);

     $('#memory_depth').val(extension_settings.memory.depth).trigger('input');
     $('#memory_role').val(extension_settings.memory.role).trigger('input');
     $('#memory_summary_prompt_role').val(extension_settings.memory.summaryPromptRole).trigger('input');
+    initSummaryConnectionProfileDropdown();
     $(`input[name="memory_position"][value="${extension_settings.memory.position}"]`).prop('checked', true).trigger('input');
     $('#memory_prompt_words_force').val(extension_settings.memory.promptForceWords).trigger('input');
     $(`input[name="memory_prompt_builder"][value="${extension_settings.memory.prompt_builder}"]`).prop('checked', true).trigger('input');

+    }
+}
+/**
+ * Resolve the maximum response length (in tokens) to request from a connection profile.
+ * Uses the configured override when it is a positive number; otherwise falls back to a
+ * sensible default cap for a summary.
+ * @returns {number} Maximum tokens for the summary response
+ */
+function getSummaryProfileMaxTokens() {
+    const override = Number(extension_settings.memory.overrideResponseLength);
+    return override > 0 ? override : 1024;
+}
+/**
+ * Generate a summary using a dedicated Connection Manager profile instead of the active model.
+ * Assembles the chat transcript (via getRawSummaryPrompt) together with the summary instruction
+ * so the chosen model actually receives the chat to summarize. Honors the configured summary
+ * prompt role by delivering the instruction as a SYSTEM or USER message.
+ * @param {object} context ST context
+ * @param {string} profileId Connection Manager profile id
+ * @param {string} prompt Summary instruction
+ * @returns {Promise<{summary: string, index: number}>} Generated summary and the index it covers up to
+ */
+async function summarizeWithConnectionProfile(context, profileId, prompt) {
+    const { rawPrompt, lastUsedIndex } = await getRawSummaryPrompt(context, prompt);
+    if (lastUsedIndex === null || lastUsedIndex === -1) {
+        // Nothing to summarize; mirror the raw builder's guard.
+        return { summary: '', index: lastUsedIndex };
+    }
+    const useUserRole = getSummaryPromptRole() === extension_prompt_roles.USER;
+    // sendRequest accepts an array of chat-completion messages, so deliver the instruction
+    // as a separate SYSTEM or USER message (per the configured role) with the transcript as
+    // a USER message. For text-completion profiles these are concatenated downstream.
+    const messages = useUserRole
+        ? [{ role: 'user', content: [prompt, rawPrompt].filter(x => x).join('\n\n') }]
+        : [{ role: 'system', content: prompt }, { role: 'user', content: rawPrompt }];
+    const filteredMessages = messages.filter(m => m.content);
+    const data = await ConnectionManagerRequestService.sendRequest(profileId, filteredMessages, getSummaryProfileMaxTokens());
+    // Non-streaming requests resolve to ExtractedData ({ content, reasoning }); extract the plain text.
+    const text = typeof data === 'object' && data !== null ? String(data.content ?? '') : '';
+    return { summary: removeReasoningFromString(text), index: lastUsedIndex };
+}
 async function summarizeChatMain(context, force, skipWIAN) {
     const prompt = await getSummaryPromptForNow(context, force);

     let summary = '';
     let index = null;
+    // A dedicated connection profile takes precedence over the DEFAULT/RAW builder distinction:
+    // it always assembles the transcript + instruction and routes through Connection Manager.
+    const summaryConnectionProfile = extension_settings.memory.summaryConnectionProfile;
+    if (summaryConnectionProfile) {
+        let profileSucceeded = false;
+        try {
+            inApiCall = true;
+            const result = await summarizeWithConnectionProfile(context, summaryConnectionProfile, prompt);
+            if (result.index === null || result.index === -1) {
+                if (force) {
+                    toastr.info('To try again, remove the latest summary.', 'No messages found to summarize');
+                }
+                return null;
+            }
+            summary = result.summary;
+            index = result.index;
+            profileSucceeded = true;
+        } catch (error) {
+            // Connection Manager disabled, unsupported API, deleted profile, etc.
+            // Fall back to the regular (active model) path below so summarization still works.
+            console.warn('Summarize: connection profile failed, falling back to active model', error);
+        } finally {
+            inApiCall = false;
+        }
+        if (profileSucceeded) {
+            if (!summary) {
+                console.warn('Empty summary received');
+                return;
+            }
+            if (isContextChanged(context)) {
+                return;
+            }
+            setMemoryContext(summary, true, index);
+            return summary;
+        }
+    }
     if (prompt_builders.DEFAULT === extension_settings.memory.prompt_builder) {
         try {
             inApiCall = true;

public/scripts/extensions/memory/settings.html+3 -0 Viewed

                             <option value="1" data-i18n="User">User</option>
                         </select>
                         <small data-i18n="ext_sum_prompt_role_desc">The role used to send the summary request instruction to the model.</small>
+                        <label for="memory_summary_connection_profile" data-i18n="ext_sum_connection_profile">Summary Connection Profile</label>
+                        <select id="memory_summary_connection_profile" class="text_pole"></select>
+                        <small data-i18n="ext_sum_connection_profile_desc">Summarize with a specific connection profile instead of the active model. Leave empty to use the active model.</small>
                         <label for="memory_prompt_words"><span data-i18n="ext_sum_target_length_1">Target summary length</span> <span data-i18n="ext_sum_target_length_2">(</span><span id="memory_prompt_words_value"></span><span data-i18n="ext_sum_target_length_3"> words)</span></label>
                         <input id="memory_prompt_words" type="range" value="{{defaultSettings.promptWords}}" min="{{defaultSettings.promptMinWords}}" max="{{defaultSettings.promptMaxWords}}" step="{{defaultSettings.promptWordsStep}}" />
                         <label for="memory_override_response_length">