03bd47c3 · permissionBRICK/SillyTavern

     extension_prompt_types,
     generateQuietPrompt,
     is_send_press,
+    online_status,
     saveSettingsDebounced,
     substituteParamsExtended,
     generateRaw,

+    }
+}
-/**
+// Warn at most once per session if a summary connection profile is selected but cannot be
- * Resolve the maximum response length (in tokens) to request from a connection profile.
+// applied because there is no active connection profile to restore afterward.
- * Uses the configured override when it is a positive number; otherwise falls back to a
+let summaryProfileWarnedNoBaseProfile = false;
- * 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.
+ * Runs a callback with a specific Connection Manager profile temporarily active, then restores
- * Assembles the chat transcript (via getRawSummaryPrompt) together with the summary instruction
+ * the previously active profile. This lets the summary be generated by the chosen LLM using the
- * so the chosen model actually receives the chat to summarize. Honors the configured summary
+ * full summarization pipeline (generateQuietPrompt/generateRaw) under that connection, instead of
- * prompt role by delivering the instruction as a SYSTEM or USER message.
+ * a context-free Connection Manager request that some providers run with the wrong model.
- * @param {object} context ST context
+ *
- * @param {string} profileId Connection Manager profile id
+ * The switch is only performed when a real connection profile is currently active (so it can be
- * @param {string} prompt Summary instruction
+ * reliably restored). When none is active — i.e. the user drives the API panel manually —
- * @returns {Promise<{summary: string, index: number}>} Generated summary and the index it covers up to
+ * switching to a profile could not be undone without clobbering those manual settings, so we leave
+ * the active model in place and warn once.
+ *
+ * @param {string} targetProfileId Profile to activate for the duration of the callback.
+ * @param {() => Promise<any>} callback Work to run while the target profile is active.
+ * @returns {Promise<any>} The callback's result.
  */
-async function summarizeWithConnectionProfile(context, profileId, prompt) {
+async function withConnectionProfile(targetProfileId, callback) {
-    const { rawPrompt, lastUsedIndex } = await getRawSummaryPrompt(context, prompt);
+    const select = /** @type {HTMLSelectElement} */ (document.getElementById('connection_profiles'));
+    const connectionManager = extension_settings.connectionManager;
+    const currentProfileId = connectionManager?.selectedProfile;
+    const canSwitch = !!select
+        && !!connectionManager
+        && Array.isArray(connectionManager.profiles)
+        && connectionManager.profiles.some(p => p.id === targetProfileId)
+        && Array.from(select.options).some(o => o.value === targetProfileId)
+        && !!currentProfileId // a real profile is active, so it can be restored afterwards
+        && currentProfileId !== targetProfileId;
+    if (!canSwitch) {
+        // Selected but no base profile to restore from -> use the active model and warn once.
+        if (targetProfileId && !currentProfileId && !summaryProfileWarnedNoBaseProfile) {
+            summaryProfileWarnedNoBaseProfile = true;
+            toastr.info('The summary connection profile is only applied while a connection profile is active (so the original can be restored). Using the current model.', 'Summarize');
+        }
+        return await callback();
+    }
+    const switchToProfile = async (profileId) => {
+        const loaded = new Promise(resolve => eventSource.once(event_types.CONNECTION_PROFILE_LOADED, resolve));
+        const timeout = new Promise(resolve => setTimeout(resolve, 10000));
+        const index = Array.from(select.options).findIndex(o => o.value === profileId);
+        select.selectedIndex = index >= 0 ? index : 0;
+        select.dispatchEvent(new Event('change'));
+        // Wait for the profile's commands to finish applying (don't hang forever if the event never fires).
+        await Promise.race([loaded, timeout]);
+        // Applying a profile reconnects the API asynchronously; generating before it is
+        // re-established fails instantly, so wait for the connection to come back up
+        // (mirrors the built-in /profile command). rejectOnTimeout:false -> proceed anyway after the timeout.
+        await waitUntilCondition(() => online_status !== 'no_connection', 10000, 100, { rejectOnTimeout: false });
+    };
-    if (lastUsedIndex === null || lastUsedIndex === -1) {
+    await switchToProfile(targetProfileId);
-        // Nothing to summarize; mirror the raw builder's guard.
+    try {
-        return { summary: '', index: lastUsedIndex };
+        return await callback();
+    } finally {
+        try {
+            await switchToProfile(currentProfileId);
+        } catch (err) {
+            console.error('Summarize: failed to restore the previous connection profile after summarization', err);
+        }
+    }
-    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) {

+    }
     console.log('sending summary prompt');
+    // Runs the configured summary builder (DEFAULT or RAW) against whatever connection is
+    // currently active. Returns { summary, index }, or null when there is nothing to summarize.
+    const runConfiguredSummary = async () => {
         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;
             // generateQuietPrompt always injects the instruction as a SYSTEM message
             // (the QUIET_PROMPT extension prompt has no exposed role parameter).
             // When a USER role is requested, assemble the chat transcript ourselves

@@ -859,15 +843,11 @@ async function summarizeChatMain(context, force, skipWIAN) {
859	};	843	};
860	summary = await generateQuietPrompt(params);	844	summary = await generateQuietPrompt(params);
861	}	845	}
862	} finally {
863	inApiCall = false;
864	}
865	}	846	}
866		847
867	if ([prompt_builders.RAW_BLOCKING, prompt_builders.RAW_NON_BLOCKING].includes(extension_settings.memory.prompt_builder)) {	848	if ([prompt_builders.RAW_BLOCKING, prompt_builders.RAW_NON_BLOCKING].includes(extension_settings.memory.prompt_builder)) {
868	const lock = extension_settings.memory.prompt_builder === prompt_builders.RAW_BLOCKING;	849	const lock = extension_settings.memory.prompt_builder === prompt_builders.RAW_BLOCKING;
869	try {	850	try {
870	inApiCall = true;
871	if (lock) {	851	if (lock) {
872	deactivateSendButtons();	852	deactivateSendButtons();
873	}	853	}

                 summary = removeReasoningFromString(rawSummary);
                 index = lastUsedIndex;
             } finally {
-            inApiCall = false;
                 if (lock) {
                     activateSendButtons();
+                }
+            }
+        }
+        return { summary, index };
+    };
+    // A dedicated connection profile generates the summary with the chosen LLM by temporarily
+    // switching the active connection profile (so the full summarization pipeline runs under it),
+    // then restoring the previous profile. Falls back to the active model when no profile is set
+    // or no base profile is active to restore.
+    const summaryConnectionProfile = extension_settings.memory.summaryConnectionProfile;
+    let result;
+    try {
+        inApiCall = true;
+        result = summaryConnectionProfile
+            ? await withConnectionProfile(summaryConnectionProfile, runConfiguredSummary)
+            : await runConfiguredSummary();
+    } finally {
+        inApiCall = false;
+    }
+    if (result === null) {
+        return null;
+    }
+    const { summary, index } = result;
     if (!summary) {
         console.warn('Empty summary received');
         return;

Summarize: generate under the chosen profile via profile-switch + reconnect-wait (replaces sendRequest, which ignored the model)