add response versioning support in messages for regeneration improvements

public/app/regeneration.js

@@ -0,0 +1,52 @@
+import {stream} from "./api.js";
+import {updateReasoning, finishReasoning} from "./reasoning.js";
+import {appendSources} from "./webSearch.js";
+
+// Keep the original nodes (including action handlers) until a committed answer arrives.
+export async function regenerateInline({card, scroller, endpoint, payload, renderContent, complete}) {
+    const original = [...card.childNodes];
+    const status = document.createElement("div");
+    status.className = "loading-shell";
+    status.setAttribute("role", "status");
+    status.textContent = "Generating...";
+    const body = document.createElement("div");
+    body.className = "msg-content streaming-content";
+    let committed = false;
+    const update = operation => {
+        const bottom = card.getBoundingClientRect().bottom;
+        const viewport = scroller.getBoundingClientRect();
+        const follow = bottom >= viewport.top && Math.abs(bottom - viewport.bottom) < 100;
+        operation();
+        if (follow) scroller.scrollTop += card.getBoundingClientRect().bottom - bottom;
+    };
+    // Anchor to this response, including when a long old answer collapses.
+    update(() => card.replaceChildren(original[0].cloneNode(true), status, body));
+    card.setAttribute("aria-busy", "true");
+    try {
+        await stream(endpoint, payload, {
+            onStatus: data => update(() => {
+                status.textContent = data.status === "searching" ? "Searching the web..."
+                    : data.status === "analyzing" ? "Analyzing attachments..." : "Generating...";
+                if ("sources" in data) {
+                    card.querySelector(".msg-sources")?.remove();
+                    appendSources(card, data.sources);
+                }
+            }),
+            onThinking: ({fullThinking}) => update(() => updateReasoning(card, fullThinking, true)),
+            onChunk: ({fullReply}) => update(() => {
+                status.hidden = Boolean(fullReply);
+                if (fullReply) finishReasoning(card);
+                renderContent(body, fullReply);
+            }),
+            onError: ({error}) => { throw new Error(error || "Unable to regenerate message."); },
+            onDone: ({message}) => {
+                if (!message) throw new Error("Missing regenerated response.");
+                update(() => complete(message));
+                committed = true;
+            }
+        });
+    } finally {
+        if (!committed) update(() => card.replaceChildren(...original));
+        card.removeAttribute("aria-busy");
+    }
+}

public/script.js

@@ -1,4 +1,5 @@
 import {del, get, post, put, stream} from "./app/api.js?v=20260909-account-controls";
+import {regenerateInline} from "./app/regeneration.js";
 import {appendSources, createWebSearchToggle} from "./app/webSearch.js";
 import {appendAttachments, createAttachmentComposer} from "./app/attachments.js";
 import {createReasoningToggle, finishReasoning, updateReasoning} from "./app/reasoning.js?v=20260921-reasoning-capability";
@@ -110,6 +111,7 @@ const webSearchToggle = createWebSearchToggle(webSearchButton);
 const reasoningToggle = createReasoningToggle(document.getElementById("reasoning"));
 
 let isProcessing = false, activeChatId = null, editingPersonaId = null, editingPersonaType = "assistant", currentUsername = "";
+let regenerationChatId = null;
 const attachmentComposer = createAttachmentComposer({onChange: updateSendState, onError: message => setNotice(message, "error")});
 let chatSessions = [], currentMessages = [], assistantPersonas = [], userPersonas = [];
 let activeUserPersonaId = null, currentSummary = null, publishedPersonaIds = new Set();
@@ -568,45 +570,70 @@ async function deleteChatMessage(messageId, fallbackIndex = -1) {
 }
 
 async function retryLatestAssistantMessage(messageId, fallbackIndex = -1) {
+    if (isProcessing) return;
     const target = await resolveMessageTarget(messageId, fallbackIndex);
+    if (isProcessing) return;
     if (!target) return setNotice("Message is not ready yet. Try again.", "error");
     if (target.message.role !== "bot") return setNotice("Only assistant messages can be regenerated.", "error");
-    const retriesUsed = Math.max(0, Number(target.message.siblingCount || 1) - 1);
+    const retriesUsed = Math.max(0, Number(target.message.versionCount || 1) - 1);
     if (retriesUsed >= 5) {
         return setNotice("Retry limit reached (5).", "error");
     }
-    setLoadingState(true, {
+    const chatId = activeChatId;
-        eyebrow: "Regenerating",
+    const card = messagesDiv.querySelector(`[data-message-id="${target.message.id}"]`);
-        title: "Generating another version",
+    if (!card) return;
-        detail: "The assistant is rewriting the latest reply."
+    const payload = {model: modelSelect.value, webSearch: webSearchToggle.enabled, reasoning: reasoningToggle.enabled};
-    });
+    regenerationChatId = chatId;
-    setNotice("Regenerating reply...");
+    setLoadingState(true, false);
     try {
-        const endpoint = getMessageEndpoint(target, "/retry");
+        await regenerateInline({
-        if (!endpoint) return setNotice("Message is not ready yet. Try again.", "error");
+            card, scroller: messagesDiv, endpoint: `/chats/${chatId}/messages/${target.message.id}/retry/stream`, payload,
-        const style = await promptPopup({
+            renderContent: setMessageContent,
-            eyebrow: "Retry style",
+            complete: message => {
-            title: "Retry modifier",
+                if (activeChatId !== chatId) return;
-            description: "Leave empty for a normal retry, or use shorter, direct, emotional, stay-in-character, or dialogue.",
+                const index = getMessageIndexById(message.id);
-            label: "Style",
+                if (index < 0) return;
-            value: "",
+                currentMessages[index] = {...currentMessages[index], ...message, thinking: message.thinking, sources: message.sources};
-            placeholder: "shorter",
+                syncRetryStateFromMessage(currentMessages[index]);
-            confirmLabel: "Retry"
+                const rendered = addMessage(currentMessages[index], false, false, {messageIndex: index, detached: true});
+                card.replaceChildren(...rendered.childNodes);
+            }
         });
-        const res = await post(endpoint, {model: modelSelect.value, style: String(style || "").trim()});
+        webSearchToggle.reset();
-        if (res.error || !res.message?.content) {
+        reasoningToggle.reset();
-            return setNotice(res.error || "Unable to retry message.", "error");
+        setNotice("Reply regenerated.", "success");
-        }
+    } catch (error) {
-        const updateIndex = target.message?.id ? getMessageIndexById(target.message.id) : target.index;
+        setNotice(error.message || "Unable to regenerate message.", "error");
-        if (updateIndex >= 0) {
-            currentMessages[updateIndex] = {...currentMessages[updateIndex], ...res.message};
-            syncRetryStateFromMessage(currentMessages[updateIndex]);
-        }
-        await setActiveChat(activeChatId);
-        setNotice(`Reply regenerated (${retriesUsed + 1}/5).`, "success");
     } finally {
+        regenerationChatId = null;
         setLoadingState(false);
-        renderMessages();
+    }
+}
+
+async function switchResponseVersion(messageId, direction, fallbackIndex = -1) {
+    const target = await resolveMessageTarget(messageId, fallbackIndex);
+    if (!target) return setNotice("Message is not ready yet. Try again.", "error");
+    if (Number(target.message.versionCount) > 1) {
+        const chatId = activeChatId;
+        const card = messagesDiv.querySelector(`[data-message-id="${messageId}"]`);
+        if (!card || card.getAttribute("aria-busy") === "true") return;
+        const buttons = [...card.querySelectorAll("button")].map(button => [button, button.disabled]);
+        card.setAttribute("aria-busy", "true");
+        buttons.forEach(([button]) => { button.disabled = true; });
+        try {
+            const res = await post(`/chats/${chatId}/messages/${messageId}/versions/${target.message.versionIndex + direction}`, {});
+            if (res.error) return setNotice(res.error, "error");
+            if (activeChatId !== chatId) return;
+            const index = getMessageIndexById(messageId);
+            if (index < 0) return;
+            currentMessages[index] = {...currentMessages[index], ...res.message, thinking: res.message.thinking, sources: res.message.sources};
+            syncRetryStateFromMessage(currentMessages[index]);
+            const rendered = addMessage(currentMessages[index], false, false, {messageIndex: index, detached: true});
+            card.replaceChildren(...rendered.childNodes);
+        } finally {
+            card.removeAttribute("aria-busy");
+            buttons.forEach(([button, disabled]) => { button.disabled = disabled; });
+        }
     }
 }
 
@@ -664,6 +691,7 @@ function addMessage(contentOrMessage, isUser = false, isLoading = false, options
             ? contentOrMessage
             : {id: null, role: isUser ? "user" : "bot", content: String(contentOrMessage || "")};
         const isMessageUser = message.role === "user";
+        msgDiv.dataset.messageId = message.id;
         msgDiv.className = `msg ${isMessageUser ? "user" : "bot"}`;
         msgDiv.setAttribute("aria-label", isMessageUser ? "Your message" : "Assistant message");
         const header = document.createElement("div"), body = document.createElement("div"), actions = document.createElement("div");
@@ -702,7 +730,7 @@ function addMessage(contentOrMessage, isUser = false, isLoading = false, options
         actions.appendChild(branchBtn);
 
         if (message.role === "bot") {
-            const retriesUsed = Math.max(0, Number(message.siblingCount || 1) - 1);
+            const retriesUsed = Math.max(0, Number(message.versionCount || 1) - 1);
             const retryBtn = document.createElement("button");
             retryBtn.type = "button";
             retryBtn.className = "msg-action-btn";
@@ -717,30 +745,32 @@ function addMessage(contentOrMessage, isUser = false, isLoading = false, options
             actions.appendChild(retryBtn);
         }
 
-        if (Number(message.siblingCount || 1) > 1) {
+        for (const versioned of [true, false]) {
+                if (Number((versioned ? message.versionCount : message.siblingCount) || 1) <= 1) continue;
+                const navigate = versioned ? switchResponseVersion : switchRetryVariant;
                 const prevBtn = document.createElement("button");
                 prevBtn.type = "button";
                 prevBtn.className = "msg-action-btn";
                 prevBtn.textContent = "Prev";
-                prevBtn.setAttribute("aria-label", "Show previous retry variant");
+                prevBtn.setAttribute("aria-label", versioned ? "Show previous response version" : "Show previous branch");
-                prevBtn.disabled = !message.previousSiblingId;
+                prevBtn.disabled = versioned ? message.versionIndex <= 0 : !message.previousSiblingId;
                 prevBtn.addEventListener("click", () => {
-                    void switchRetryVariant(message.id, -1, options.messageIndex ?? -1);
+                    void navigate(message.id, -1, options.messageIndex ?? -1);
                 });
 
                 const nextBtn = document.createElement("button");
                 nextBtn.type = "button";
                 nextBtn.className = "msg-action-btn";
                 nextBtn.textContent = "Next";
-                nextBtn.setAttribute("aria-label", "Show next retry variant");
+                nextBtn.setAttribute("aria-label", versioned ? "Show next response version" : "Show next branch");
-                nextBtn.disabled = !message.nextSiblingId;
+                nextBtn.disabled = versioned ? message.versionIndex >= message.versionCount - 1 : !message.nextSiblingId;
                 nextBtn.addEventListener("click", () => {
-                    void switchRetryVariant(message.id, 1, options.messageIndex ?? -1);
+                    void navigate(message.id, 1, options.messageIndex ?? -1);
                 });
 
                 const indexBadge = document.createElement("span");
                 indexBadge.className = "msg-variant-index";
-                indexBadge.textContent = `${Number(message.siblingIndex || 0) + 1}/${message.siblingCount}`;
+                indexBadge.textContent = versioned ? `${message.versionIndex + 1} / ${message.versionCount}` : `${Number(message.siblingIndex || 0) + 1}/${message.siblingCount}`;
                 actions.append(prevBtn, indexBadge, nextBtn);
         }
 
@@ -759,8 +789,10 @@ function addMessage(contentOrMessage, isUser = false, isLoading = false, options
             msgDiv.appendChild(actions);
         }
     }
-    messagesDiv.appendChild(msgDiv);
+    if (!options.detached) {
-    scrollMessagesToBottom();
+        messagesDiv.appendChild(msgDiv);
+        scrollMessagesToBottom();
+    }
     return msgDiv;
 }
 
@@ -1388,8 +1420,9 @@ function setLoadingState(loading, overlayOptions = null) {
 }
 
 async function stopGeneration() {
-    if (!isProcessing || !activeChatId) return;
+    const chatId = regenerationChatId ?? activeChatId;
-    const res = await post(`/chats/${activeChatId}/stop`, {});
+    if (!isProcessing || !chatId) return;
+    const res = await post(`/chats/${chatId}/stop`, {});
     if (res.error) return setNotice(res.error, "error");
     setNotice(res.stopped ? "Stopping generation..." : "Generation already finished.");
 }

scripts/lint.mjs

@@ -35,6 +35,7 @@ const files = [
     "public/settings.js",
     "public/market.js",
     "public/app/api.js",
+    "public/app/regeneration.js",
     "public/app/webSearch.js",
     "public/app/reasoning.js",
     "public/app/modelCapabilities.js",

src/config.js

@@ -58,7 +58,7 @@ export function createConfig(overrides = {}) {
         chatSummaryMaxChars: int(process.env.CHAT_SUMMARY_MAX_CHARS, 2200),
         chatMaxRetries: int(process.env.CHAT_MAX_RETRIES, 5),
         modelApiBaseUrl: process.env.MODEL_API_BASE_URL || "https://ai.krishd.ch/api",
-        modelRequestTimeoutMs: int(process.env.MODEL_REQUEST_TIMEOUT_MS, 180000),
+        modelRequestYTimeoutMs: int(process.env.MODEL_REQUEST_TIMEOUT_MS, 180000),
         modelConnectTimeoutMs: Math.max(1, int(process.env.MODEL_CONNECT_TIMEOUT_MS, 30000)),
         modelInactivityTimeoutMs: Math.max(1, int(process.env.MODEL_INACTIVITY_TIMEOUT_MS, int(process.env.MODEL_REQUEST_TIMEOUT_MS, 180000))),
         modelMaxGenerationMs: Math.max(1, int(process.env.MODEL_MAX_GENERATION_MS, 600000)),

src/db/repositories.js

@@ -871,6 +871,10 @@ export function createRepositories(db, config) {
                 if (message.sourceId != null) importedMessageIds.set(message.sourceId, Number(result.lastInsertRowid));
                 if (message.sources) statements.updateChatMessageSources.run(JSON.stringify(message.sources), result.lastInsertRowid, chatId);
                 if (message.thinking) statements.updateChatMessageThinking.run(message.thinking, result.lastInsertRowid, chatId);
+                for (const version of message.responseVersions || []) {
+                    db.prepare("INSERT INTO chat_message_versions (message_id, content, thinking, sources, model_name, created_at) VALUES (?, ?, ?, ?, ?, COALESCE(?, CURRENT_TIMESTAMP))")
+                        .run(result.lastInsertRowid, version.content, version.thinking, JSON.stringify(version.sources), version.modelName, version.createdAt);
+                }
                 for (const [position, attachment] of (message.attachments || []).slice(0, config.uploadMaxFiles || 5).entries()) {
                     // JSON backups contain references, not files. Only restore existing, owned images.
                     db.prepare("INSERT OR IGNORE INTO message_attachments(message_id,attachment_id,position) SELECT ?,id,? FROM attachments WHERE id=? AND username=? AND status='attached'")
@@ -1046,6 +1050,29 @@ export function createRepositories(db, config) {
             username
         ),
         updateChatPreferredModel: (chatId, username, model) => statements.updateChatPreferredModel.run(model || null, chatId, username),
+        listMessageVersions: (messageId) => db.prepare("SELECT * FROM chat_message_versions WHERE message_id = ? ORDER BY id").all(messageId),
+        saveMessageVersion: (chatId, messageId, version = null, selectedIndex = null) => db.transaction(() => {
+            const row = statements.getChatMessageById.get(messageId, chatId);
+            if (!row) throw new Error("Message not found");
+            let versions = db.prepare("SELECT * FROM chat_message_versions WHERE message_id = ? ORDER BY id").all(messageId);
+            const insert = db.prepare("INSERT INTO chat_message_versions (message_id, content, thinking, sources, model_name) VALUES (?, ?, ?, ?, ?)");
+            if (!versions.length) {
+                db.prepare("INSERT INTO chat_message_versions (message_id, content, thinking, sources, model_name, created_at) VALUES (?, ?, ?, ?, ?, ?)")
+                    .run(messageId, row.content, row.thinking, row.sources, row.model_name, row.created_at);
+            } else {
+                // Edits keep their existing semantics: update the selected answer.
+                db.prepare("UPDATE chat_message_versions SET content = ?, thinking = ?, sources = ?, model_name = ? WHERE id = ?")
+                    .run(row.content, row.thinking, row.sources, row.model_name, versions[row.retry_active_index]?.id ?? versions[0].id);
+            }
+            if (version) insert.run(messageId, version.content, version.thinking || null, JSON.stringify(version.sources ?? null), version.modelName);
+            versions = db.prepare("SELECT * FROM chat_message_versions WHERE message_id = ? ORDER BY id").all(messageId);
+            const index = version ? versions.length - 1 : selectedIndex;
+            if (!Number.isInteger(index) || !versions[index]) throw new Error("Version is out of range");
+            const active = versions[index];
+            db.prepare(`UPDATE chat_messages SET content = ?, thinking = ?, sources = ?, model_name = ?,
+                retry_variants = ?, retry_active_index = ?, retry_retries_used = ? WHERE id = ? AND chat_id = ?`)
+                .run(active.content, active.thinking, active.sources, active.model_name, JSON.stringify(versions.map(item => item.content)), index, versions.length - 1, messageId, chatId);
+        })(),
         updateChatMessage: (chatId, messageId, content) => statements.updateChatMessage.run(content, messageId, chatId),
         updateChatMessageRetryState: (chatId, messageId, payload) => statements.updateChatMessageWithRetryState.run(
             payload.content,
@@ -1278,6 +1305,10 @@ export function createRepositories(db, config) {
                     persona_state: this.getPersonaChatState(chat.id, username),
                     messages: this.listChatMessages(chat.id).map((message) => ({
                         ...message,
+                        responseVersions: this.listMessageVersions(message.id).map(version => ({
+                            content: version.content, thinking: version.thinking,
+                            sources: parseJsonArray(version.sources), modelName: version.model_name, createdAt: version.created_at
+                        })),
                         attachments: db.prepare("SELECT a.* FROM attachments a JOIN message_attachments m ON a.id=m.attachment_id WHERE m.message_id=? ORDER BY m.position").all(message.id).map(attachmentMetadata),
                         retryVariants: parseJsonArray(message.retry_variants)
                     }))

src/db/schema.js

@@ -236,6 +236,16 @@ export function initializeSchema(db) {
 }
 
 export function runMigrations(db, {dropLegacyChats = true} = {}) {
+    db.exec(`CREATE TABLE IF NOT EXISTS chat_message_versions (
+        id INTEGER PRIMARY KEY AUTOINCREMENT,
+        message_id INTEGER NOT NULL REFERENCES chat_messages(id) ON DELETE CASCADE,
+        content TEXT NOT NULL,
+        thinking TEXT,
+        sources TEXT,
+        model_name TEXT,
+        created_at DATETIME DEFAULT CURRENT_TIMESTAMP
+    );
+    CREATE INDEX IF NOT EXISTS chat_message_versions_message ON chat_message_versions(message_id);`);
     const ensureColumn = (tableName, columns, name, sql) => {
         if (!columns.some((column) => column.name === name)) {
             db.prepare(sql).run();

src/routes/chats.js

@@ -105,6 +105,14 @@ const preferredModelValidator = (body) => {
 export function createChatsRouter({repositories, chatService, modelService, config, chatRateLimiters}) {
     const router = express.Router();
     router.use(requireLogin);
+    router.use(["/chats/:id/messages/by-index/:index", "/chats/:id/messages/:messageId"], (req, res, next) => {
+        const chatId = Number(req.params.id);
+        const messageId = req.params.messageId ? Number(req.params.messageId) : chatService.getChatMessageByIndex(chatId, req.params.index)?.id;
+        if (["PUT", "DELETE", "POST"].includes(req.method) && chatService.isRegeneratingMessage(chatId, messageId)) {
+            return res.status(409).json({error: "This response is being regenerated. Wait for it to finish."});
+        }
+        next();
+    });
 
     router.get("/models", async (req, res) => {
         try {
@@ -475,7 +483,21 @@ export function createChatsRouter({repositories, chatService, modelService, conf
         }
     });
 
-    router.post("/chats/:id/messages/:messageId/retry", ...chatRateLimiters, async (req, res) => {
+    router.post("/chats/:id/messages/:messageId/versions/:index", (req, res) => {
+        const chatId = Number(req.params.id), messageId = Number(req.params.messageId);
+        if (!repositories.getChat(chatId, req.session.user)) return res.status(404).json({error: "Chat not found"});
+        const message = repositories.getChatMessage(chatId, messageId);
+        if (!message || message.role !== "bot") return res.status(404).json({error: "Assistant message not found"});
+        if (chatService.isGenerating(chatId)) return res.status(409).json({error: "Wait for generation to finish"});
+        const index = Number(req.params.index);
+        const versions = repositories.listMessageVersions(messageId);
+        if (!Number.isInteger(index) || index < 0 || index >= versions.length) return res.status(400).json({error: "Version is out of range"});
+        repositories.saveMessageVersion(chatId, messageId, null, index);
+        repositories.updateChatContextSummary(chatId, req.session.user, null, 0);
+        res.json({message: chatService.formatChatMessage(repositories.getChatMessage(chatId, messageId))});
+    });
+
+    router.post(["/chats/:id/messages/:messageId/retry", "/chats/:id/messages/:messageId/retry/stream"], ...chatRateLimiters, async (req, res) => {
         const chatId = Number(req.params.id);
         const messageId = Number(req.params.messageId);
         const chat = repositories.getChat(chatId, req.session.user);
@@ -484,19 +506,45 @@ export function createChatsRouter({repositories, chatService, modelService, conf
         if (!targetMessage) return res.status(404).json({error: "Message not found"});
         if (targetMessage.role !== "bot") return res.status(400).json({error: "Only assistant messages can be regenerated"});
 
+        const streaming = req.path.endsWith("/stream");
+        const controller = new AbortController();
+        let finished = false;
+        const sendEvent = (event, data) => {
+            if (!res.destroyed) res.write(`event: ${event}\ndata: ${JSON.stringify(data)}\n\n`);
+        };
+        if (streaming) {
+            res.once("close", () => { if (!finished) controller.abort(); });
+            res.writeHead(200, {"Content-Type": "text/event-stream", "Cache-Control": "no-cache, no-transform", Connection: "keep-alive", "X-Accel-Buffering": "no"});
+        }
         try {
             const result = await chatService.retryMessage({
                 user: req.session.user,
                 chatId,
                 selectedModel: String(req.body?.model || "qwen3:14b"),
                 targetMessage,
-                retryStyle: String(req.body?.style || "").trim() || null
+                retryStyle: String(req.body?.style || "").trim() || null,
+                webSearch: req.body?.webSearch === true,
+                reasoning: req.body?.reasoning === true,
+                signal: controller.signal,
+                ...(streaming ? {
+                    onChunk: (chunk, fullReply) => sendEvent("chunk", {chunk, fullReply}),
+                    onThinking: (chunk, fullThinking) => sendEvent("thinking", {chunk, fullThinking}),
+                    onStatus: status => sendEvent("status", status)
+                } : {})
             });
+            if (streaming) {
+                sendEvent(result.error ? "error" : "done", result.error ? {error: result.error, searchState: result.searchState} : {...result, reply: result.message.content});
+                return;
+            }
             if (result.error) return res.status(result.status).json({error: result.error});
             return res.json(result);
         } catch (error) {
             const mapped = modelService.mapError(error);
+            if (streaming) { sendEvent("error", mapped.body); return; }
             return res.status(mapped.status).json(mapped.body);
+        } finally {
+            finished = true;
+            if (streaming) res.end();
         }
     });
 
@@ -514,7 +562,9 @@ export function createChatsRouter({repositories, chatService, modelService, conf
                 chatId,
                 selectedModel: String(req.body?.model || "qwen3:14b"),
                 targetMessage,
-                retryStyle: String(req.body?.style || "").trim() || null
+                retryStyle: String(req.body?.style || "").trim() || null,
+                webSearch: req.body?.webSearch === true,
+                reasoning: req.body?.reasoning === true
             });
             if (result.error) return res.status(result.status).json({error: result.error});
             return res.json(result);

src/services/chatService.js

@@ -24,12 +24,22 @@ import {supportsReasoning} from "../../public/app/modelCapabilities.js";
 export function createChatService(repositories, modelService, config, webSearchProvider = new SearXNGSearchProvider(config), attachmentService = null) {
     const formatChatMessage = row => {
         const message = formatBaseChatMessage(row);
+        if (message?.role === "bot") {
+            const versions = repositories.listMessageVersions(row.id);
+            message.versionCount = versions.length || 1;
+            message.versionIndex = versions.length ? Number(row.retry_active_index || 0) : 0;
+            if (versions.length) message.responseVersions = versions.map(version => ({
+                content: version.content, thinking: version.thinking,
+                sources: parseSources(version.sources), modelName: version.model_name, createdAt: version.created_at
+            }));
+        }
         const attachments = row && attachmentService?.list(row.id);
         return message && attachments?.length ? {...message, attachments} : message;
     };
     const chatGenerationQueues = new Map();
     const summaryQueues = new Map();
     const activeGenerations = new Map();
+    const regeneratingMessages = new Map();
     const maxRetries = Number(config.chatMaxRetries ?? 5);
     const now = typeof config.now === "function" ? config.now : () => new Date();
 
@@ -84,6 +94,8 @@ export function createChatService(repositories, modelService, config, webSearchP
     });
 
     const persistChatMessageRetryState = ({chatId, messageId, content, retryVariants, retryActiveIndex, retryRetriesUsed, retryPromptMessageId}) => {
+        const versioned = repositories.listMessageVersions(messageId).length > 0;
+        if (versioned) repositories.saveMessageVersion(chatId, messageId, null, retryActiveIndex);
         repositories.updateChatMessageRetryState(chatId, messageId, {
             content,
             retryVariants: retryVariants || [content || ""],
@@ -91,6 +103,7 @@ export function createChatService(repositories, modelService, config, webSearchP
             retryRetriesUsed: Number(retryRetriesUsed || 0),
             retryPromptMessageId: retryPromptMessageId ?? null
         });
+        if (versioned) repositories.saveMessageVersion(chatId, messageId, null, retryActiveIndex);
     };
 
     const getChatMessageByIndex = (chatId, index) => {
@@ -174,7 +187,7 @@ export function createChatService(repositories, modelService, config, webSearchP
         return repositories.createChat(user, persona.name, persona.id, userPersonaId, scenarioPrompt, scenarioSummary);
     };
 
-    const generateRoleplayOpener = async ({selectedModel = "qwen3:14b", assistantPersona, userPersona, scenarioPrompt, sceneSummary, onThinking = null}) => {
+    const generateRoleplayOpener = async ({selectedModel = "qwen3:14b", assistantPersona, userPersona, scenarioPrompt, sceneSummary, onThinking = null, onChunk = null, signal = null, generation = {}, reasoning = false}) => {
         const baseMessages = [
             {role: "system", content: buildPersonaPrompt(assistantPersona)},
             ...(userPersona ? [{role: "system", content: buildUserPersonaPrompt(userPersona)}] : []),
@@ -182,20 +195,24 @@ export function createChatService(repositories, modelService, config, webSearchP
             {role: "user", content: buildRoleplayOpenerPrompt({assistantPersona, userPersona, scenarioPrompt, sceneSummary})}
         ];
 
-        let opener = await modelService.generateReply(selectedModel, baseMessages, {onThinking});
+        const generate = messages => onChunk
+            ? modelService.streamReply(selectedModel, messages, onChunk, {onThinking, signal, generation, reasoning})
+            : modelService.generateReply(selectedModel, messages, {onThinking, signal, generation, reasoning});
+        let opener = await generate(baseMessages);
         if (!containsUserVoiceInRoleplayOpener(opener, userPersona)) {
             return opener;
         }
 
         onThinking?.("", "");
-        return modelService.generateReply(selectedModel, [
+        onChunk?.("", "");
+        return generate([
             ...baseMessages,
             {role: "assistant", content: opener || ""},
             {
                 role: "user",
                 content: "Rewrite this opening. Hard rule: only the assistant character may speak or act. Never write or imply user speech, thoughts, feelings, reactions, decisions, actions, or labels like 'User:'/'You:'. Stop before the user's reply."
             }
-        ], {onThinking});
+        ]);
     };
 
     const generateValidatedRoleplayReply = async ({selectedModel, messagesPayload, userPersona, signal = null, generation = {}, onThinking = null, onChunk = null, reasoning = false}) => {
@@ -285,13 +302,20 @@ export function createChatService(repositories, modelService, config, webSearchP
         return {session: activeSession, activePersona, activeUserPersona, messagesPayload};
     };
 
-    const buildRetryPayload = async ({chatId, user, session, selectedModel, targetMessage, retryStyle = null, signal = null, generation = {}}) => {
+    const buildRetryPayload = async ({chatId, user, session, selectedModel, targetMessage, retryStyle = null, signal = null, generation = {}, webSearch = false, reasoning = false, onChunk = null, onThinking = null, onStatus = null}) => {
+        let thinking = "";
+        const captureThinking = (chunk, fullThinking) => {
+            if (!reasoning) return;
+            thinking = fullThinking;
+            onThinking?.(chunk, fullThinking);
+        };
+        onStatus?.({status: "accepted"});
         const parentMessage = targetMessage.parent_message_id
             ? repositories.getChatMessage(chatId, targetMessage.parent_message_id)
             : null;
         const promptMessage = parentMessage?.role === "user"
             ? parentMessage
-            : repositories.getPreviousUserMessage(chatId, targetMessage.id);
+            : repositories.listChatBranchThroughMessage(chatId, targetMessage.id).slice(0, -1).reverse().find(entry => entry.role === "user");
         const activePersona = repositories.getAssistantPersonaForChat(chatId, user);
         const activeUserPersona = getUserPersonaForSession(chatId, user, session, activePersona);
 
@@ -299,9 +323,8 @@ export function createChatService(repositories, modelService, config, webSearchP
             if (!activePersona || !session.scenario_summary) {
                 return {error: "No previous user prompt found for retry"};
             }
-            let thinking = "";
             const opener = await generateRoleplayOpener({
-                onThinking: (_chunk, fullThinking) => { thinking = fullThinking; },
+                onThinking: captureThinking, onChunk, signal, generation, reasoning,
                 selectedModel,
                 assistantPersona: activePersona,
                 userPersona: activeUserPersona,
@@ -314,25 +337,38 @@ export function createChatService(repositories, modelService, config, webSearchP
             return {fullReply: opener, thinking, promptMessageId: null};
         }
 
+        if (attachmentService?.list(promptMessage.id).length) onStatus?.({status: "analyzing"});
         await attachmentService?.analyze(promptMessage.id, promptMessage.content, signal);
+        let sources = null;
+        let searchState = "skipped";
+        if (webSearch) {
+            const search = await retrieveWebSearch(promptMessage.content, webSearchProvider, {
+                signal,
+                referenceText: attachmentService?.searchContext(promptMessage.id) || "",
+                onSearching: () => onStatus?.({status: "searching"})
+            });
+            sources = search.sources;
+            searchState = search.state;
+        }
+        onStatus?.({status: "generating", searchState, sources});
         const {messagesPayload} = buildConversationContext({
             user,
             chatId,
             session,
             message: null,
             upToMessageId: promptMessage.id,
-            sources: targetMessage.sources ? parseSources(targetMessage.sources) : null
+            sources
         });
 
         const styledPayload = appendRetryStyleInstruction(messagesPayload, retryStyle);
-        let thinking = "";
-        const onThinking = (_chunk, fullThinking) => { thinking = fullThinking; };
         const fullReply = activePersona
-            ? await generateValidatedRoleplayReply({selectedModel, messagesPayload: styledPayload, userPersona: activeUserPersona, signal, generation, onThinking})
+            ? await generateValidatedRoleplayReply({selectedModel, messagesPayload: styledPayload, userPersona: activeUserPersona, signal, generation, onThinking: captureThinking, onChunk, reasoning})
-            : await modelService.generateReply(selectedModel, styledPayload, {signal, generation, onThinking});
+            : onChunk
+                ? await modelService.streamReply(selectedModel, styledPayload, onChunk, {signal, generation, onThinking: captureThinking, reasoning})
+                : await modelService.generateReply(selectedModel, styledPayload, {signal, generation, onThinking: captureThinking, reasoning});
 
         if (!fullReply) return {error: "Failed to regenerate message"};
-        return {fullReply, thinking, promptMessageId: promptMessage.id};
+        return {fullReply, thinking, sources, searchState, promptMessageId: promptMessage.id};
     };
 
     const generateChatTurn = ({user, chatId, message, model, attachments = [], reuseChatId = null, webSearch = false, reasoning = false, onStatus = null, onChunk = null, onThinking = null, signal = null, parentMessageId = undefined}) =>
@@ -536,8 +572,13 @@ export function createChatService(repositories, modelService, config, webSearchP
         async streamChatMessage({user, chatId, message, model, attachments = [], webSearch = false, reasoning = false, onStatus, onChunk, onThinking, signal = null}) {
             return generateChatTurn({user, chatId, message, model, attachments, webSearch, reasoning, onStatus, onChunk, onThinking, signal});
         },
-        async retryMessage({user, chatId, selectedModel, targetMessage, retryStyle = null}) {
+        isGenerating: (chatId) => chatGenerationQueues.has(chatId),
+        isRegeneratingMessage: (chatId, messageId) => regeneratingMessages.get(chatId) === messageId,
+        async retryMessage({user, chatId, selectedModel, targetMessage, retryStyle = null, webSearch = false, reasoning = false, onChunk = null, onThinking = null, onStatus = null, signal = null}) {
+            if (chatGenerationQueues.has(chatId)) return {error: "A response is already being generated", status: 409};
+            regeneratingMessages.set(chatId, targetMessage.id);
             return withChatGenerationLock(chatId, async () => {
+                await (summaryQueues.get(chatId) || Promise.resolve());
                 const session = repositories.getChat(chatId, user);
                 if (!session) return {error: "Chat not found", status: 404};
 
@@ -545,35 +586,31 @@ export function createChatService(repositories, modelService, config, webSearchP
                 const existingMessage = formatChatMessage(freshTarget);
                 if (!existingMessage) return {error: "Message not found", status: 404};
                 if (existingMessage.role !== "bot") return {error: "Only assistant messages can be regenerated", status: 400};
-                const siblings = repositories.listSiblingChatMessages(chatId, "bot", freshTarget.parent_message_id);
+                if (existingMessage.versionCount - 1 >= maxRetries) {
-                if (siblings.length - 1 >= maxRetries) {
                     return {error: `Maximum of ${maxRetries} retries reached`, status: 400};
                 }
                 const {model, generation} = getGenerationSettings(session, selectedModel);
-                return withActiveGeneration(chatId, null, async (activeSignal) => {
+                return withActiveGeneration(chatId, signal, async (activeSignal) => {
+                    activeSignal.throwIfAborted();
                     const retryResult = await buildRetryPayload({
                         chatId, user, session, selectedModel: model, targetMessage: freshTarget, retryStyle,
-                        signal: activeSignal, generation
+                        signal: activeSignal, generation, webSearch, reasoning: reasoning === true && supportsReasoning(model), onChunk, onThinking, onStatus
                     });
                     if (retryResult.error) return {error: retryResult.error, status: 400};
-                    const retryVariants = [...siblings.map((sibling) => sibling.content), retryResult.fullReply];
+                    activeSignal.throwIfAborted();
-                    const retryState = {
+                    repositories.saveMessageVersion(chatId, freshTarget.id, {
-                        retryVariants,
+                        content: retryResult.fullReply, thinking: retryResult.thinking,
-                        retryActiveIndex: retryVariants.length - 1,
+                        sources: retryResult.sources, modelName: model
-                        retryRetriesUsed: siblings.length,
+                    });
-                        retryPromptMessageId: retryResult.promptMessageId
-                    };
-                    const inserted = repositories.insertBranchChatMessage(
-                        chatId, "bot", retryResult.fullReply, retryState, model, "complete", null, freshTarget.parent_message_id
-                    );
-                    if (freshTarget.sources) repositories.updateChatMessageSources(chatId, Number(inserted.lastInsertRowid), parseSources(freshTarget.sources));
-                    if (retryResult.thinking) repositories.updateChatMessageThinking(chatId, Number(inserted.lastInsertRowid), retryResult.thinking);
                     repositories.updateChatContextSummary(chatId, user, null, 0);
                     repositories.touchChat(chatId, user);
-                    const message = formatChatMessage(repositories.getChatMessage(chatId, Number(inserted.lastInsertRowid)));
+                    const message = formatChatMessage(repositories.getChatMessage(chatId, freshTarget.id));
-                    return {message, promptMessageId: retryResult.promptMessageId, retryStyle, branchedFromMessageId: freshTarget.id};
+                    return {message, promptMessageId: retryResult.promptMessageId, retryStyle};
                 });
-            });
+            }).catch(error => {
+                if (error instanceof WebSearchError) return {error: error.message, status: error.status, searchState: "failed"};
+                throw error;
+            }).finally(() => regeneratingMessages.delete(chatId));
         },
         async editAndResendMessage({user, chatId, targetMessage, content, model, webSearch = false, onChunk = null, signal = null}) {
             if (!targetMessage || targetMessage.role !== "user") return {error: "Only user messages can be edited and resent", status: 400};

src/services/workspaceImportService.js

@@ -86,6 +86,10 @@ function validateMessage(message, chatIndex, messageIndex, errors) {
         content,
         model_name: optionalText(message.model_name || message.modelName),
         thinking: optionalText(message.thinking),
+        responseVersions: asArray(message.responseVersions).filter(version => version && typeof version.content === "string").slice(0, 100).map(version => ({
+            content: version.content, thinking: optionalText(version.thinking), sources: parseSources(version.sources),
+            modelName: optionalText(version.modelName), createdAt: optionalText(version.createdAt)
+        })),
         attachments: asArray(message.attachments).filter(item => item && typeof item.id === "string" && /^[a-f0-9-]{36}$/.test(item.id)).slice(0, 100).map(item => ({id: item.id})),
         ...(message.sources ? {sources: parseSources(message.sources)} : {}),
         retryVariants,

test/app.test.js

@@ -455,8 +455,11 @@ test("conversation branches, edit-resend, and generation settings work", async (
     });
     assert.equal(regenerated.status, 200);
     let active = await request(server.baseUrl, `/chats/${chatId}/messages`, {jar});
-    assert.equal(active.json.messages.length, 2);
+    assert.equal(active.json.messages.length, 4);
-    assert.equal(active.json.messages[1].siblingCount, 2);
+    assert.equal(active.json.messages[1].id, firstAssistant.id);
+    assert.equal(active.json.messages[1].versionCount, 2);
+    assert.equal(active.json.messages[1].siblingCount, 1);
+    await request(server.baseUrl, `/chats/${chatId}/messages/${firstAssistant.id}/versions/0`, {method: "POST", body: {}, jar});
 
     await request(server.baseUrl, `/chats/${chatId}/branches/${originalLeaf.id}/activate`, {method: "POST", body: {}, jar});
     active = await request(server.baseUrl, `/chats/${chatId}/messages`, {jar});

test/attachments.test.js

@@ -151,7 +151,11 @@ test("images feed dedicated vision, cache, streamed reasoning, reload and shared
     const exported = await (await s.api("/exports/workspace")).json();
     assert.deepEqual(exported.workspace.chats[0].messages[0].attachments, uploads);
     const botId = history.messages.at(-1).id;
-    assert.equal((await s.api(`/chats/${chatId}/messages/${botId}/retry`, {model: "qwen3:14b"})).status, 200);
+    const retryWire = await (await s.api(`/chats/${chatId}/messages/${botId}/retry/stream`, {model: "qwen3:14b", reasoning: true})).text();
+    assert.match(retryWire, /event: chunk/);
+    assert.match(retryWire, /event: thinking/);
+    assert.match(retryWire, /event: done/);
+    assert.ok(s.requests.at(-1).messages.some(v => v.content.includes("A blue screenshot")));
     assert.equal(s.requests.filter(v => v.model === "gemma3:4b").length, 2, "Retry reuses vision cache");
     const branchResponse = await s.api(`/chats/${chatId}/messages/${botId}/branch`, {confirmed: true});
     assert.equal(branchResponse.status, 200);

test/reasoning.test.js

@@ -77,6 +77,7 @@ test("thinking streams separately and survives reload, retry, branch and import"
     assert.equal(bot.content, "Answer");
     const retry = await api(`/chats/${chatId}/messages/${bot.id}/retry`, {});
     assert.equal(retry.message.thinking, undefined);
+    await api(`/chats/${chatId}/messages/${bot.id}/versions/0`, {});
     const branch = await api(`/chats/${chatId}/messages/${bot.id}/branch`, {confirmed: true});
     assert.equal(branch.messages.at(-1).thinking, bot.thinking);
     const exported = await api(`/exports/chats/${chatId}?format=json`);
@@ -125,6 +126,19 @@ test("thinking streams separately and survives reload, retry, branch and import"
     assert.match(roleplayWire, /event: done/);
     assert.ok(requests.at(-1).messages.some(message => message.content.includes("Guide")));
     assert.equal(requests.at(-1).think, true);
+    const roleplayMessages = (await api(`/chats/${roleplay.chat.id}/messages`)).messages;
+    for (const target of [roleplayMessages[0], roleplayMessages.at(-1)]) {
+        const retryResponse = await fetch(base + `/chats/${roleplay.chat.id}/messages/${target.id}/retry/stream`, {
+            method: "POST", headers: {"Content-Type": "application/json", Cookie: cookie},
+            body: JSON.stringify({model: "qwen3:14b", reasoning: true})
+        });
+        const retryWire = await retryResponse.text();
+        assert.match(retryWire, /event: thinking/);
+        assert.match(retryWire, /event: chunk/);
+        assert.match(retryWire, /event: done/);
+        assert.ok(requests.at(-1).messages.some(message => message.content.includes("Guide")));
+        assert.equal(requests.at(-1).think, true);
+    }
 });
 
 test("reasoning disclosure is collapsed, preserves open state and renders text safely", (t) => {

test/regeneration.test.js

@@ -0,0 +1,170 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+import {createServer} from "node:http";
+import {createApp} from "../src/app.js";
+import Database from "better-sqlite3";
+import {runMigrations} from "../src/db/schema.js";
+
+const deferred = () => { let resolve; const promise = new Promise(done => { resolve = done; }); return {promise, resolve}; };
+
+async function fixture(t) {
+    let mode = "success", sequence = 0, searches = 0;
+    let started = deferred(), release = deferred();
+    const calls = [];
+    const search = createServer((_req, res) => {
+        searches++;
+        if (mode === "search-fail") { res.writeHead(503); res.end("Unavailable"); return; }
+        res.setHeader("Content-Type", "application/json");
+        res.end(JSON.stringify({results: [{title: "Satellite launch news", url: `https://example.org/news/${searches}`, content: "Latest satellite launch news today: a research satellite launched successfully."}]}));
+    });
+    await new Promise(resolve => search.listen(0, "127.0.0.1", resolve));
+    const modelService = {
+        async generateReply() { return `Original ${++sequence}`; },
+        async streamReply(model, messages, chunk, options) {
+            calls.push({model, messages, options});
+            if (options.reasoning) options.onThinking("New reasoning", "New reasoning");
+            chunk("Partial", "Partial");
+            started.resolve();
+            if (mode === "hold") await new Promise((resolve, reject) => {
+                release.promise.then(resolve);
+                options.signal.addEventListener("abort", () => reject(new Error("Generation stopped")), {once: true});
+            });
+            if (mode === "fail") throw new Error("Model failed after partial output");
+            if (mode === "empty") return "";
+            chunk(" answer", "Partial answer");
+            return "Partial answer";
+        },
+        mapError(error) { return {status: 500, body: {error: error.message}}; }
+    };
+    const app = createApp({modelService, config: {dbPath: ":memory:", testMode: true, sessionSecret: "regeneration-test",
+        chatSummaryUpdateEveryMessages: 0, eventLoopLagMonitorEnabled: false, structuredRequestLoggingEnabled: false,
+        slowRequestLoggingEnabled: false, searxngUrl: `http://127.0.0.1:${search.address().port}`}});
+    const server = app.listen(0, "127.0.0.1");
+    await new Promise(resolve => server.once("listening", resolve));
+    t.after(async () => {
+        release.resolve();
+        await new Promise(resolve => { server.closeAllConnections(); server.close(resolve); });
+        await new Promise(resolve => { search.closeAllConnections(); search.close(resolve); });
+        await app.locals.close();
+    });
+    let cookie;
+    const api = async (path, body, method = body ? "POST" : "GET") => {
+        const response = await fetch(`http://127.0.0.1:${server.address().port}${path}`, {method,
+            headers: {"Content-Type": "application/json", ...(cookie ? {Cookie: cookie} : {})}, body: body ? JSON.stringify(body) : undefined});
+        if (response.headers.get("set-cookie")) cookie = response.headers.get("set-cookie").split(";")[0];
+        return response;
+    };
+    await api("/register", {username: "regenerator", password: "test-password"});
+    const chatId = (await (await api("/chats", {title: "Versions"})).json()).chat.id;
+    await api("/chat", {chatId, message: "Search the web for the latest satellite launch news today"});
+    await api("/chat", {chatId, message: "Later private question"});
+    const messages = (await (await api(`/chats/${chatId}/messages`)).json()).messages;
+    const bot = messages[1];
+    app.locals.repositories.updateChatMessageThinking(chatId, bot.id, "Original reasoning");
+    app.locals.repositories.updateChatMessageSources(chatId, bot.id, [{title: "Old", url: "https://example.org/old", snippet: "Old evidence"}]);
+    return {app, api, chatId, bot, messages, calls,
+        get searches() { return searches; },
+        path: `/chats/${chatId}/messages/${bot.id}`,
+        hold() { mode = "hold"; started = deferred(); release = deferred(); return started.promise; },
+        setMode(value) { mode = value; }, release: () => release.resolve(),
+        history: async () => (await (await api(`/chats/${chatId}/messages`)).json()).messages};
+}
+
+test("regeneration streams before commit in the same message; versions retain sources and reasoning and branch from selection", async t => {
+    const s = await fixture(t);
+    const started = s.hold();
+    const response = await s.api(`${s.path}/retry/stream`, {reasoning: true, webSearch: true});
+    await started;
+    const reader = response.body.getReader();
+    let wire = new TextDecoder().decode((await reader.read()).value);
+    while (!wire.includes("Partial")) wire += new TextDecoder().decode((await reader.read()).value);
+    assert.match(wire, /event: thinking/);
+    assert.match(wire, /Partial/);
+    assert.doesNotMatch(wire, /event: done/);
+    let history = await s.history();
+    assert.equal(history[1].content, s.bot.content, "database retains original while streaming");
+    assert.equal(history[1].versionCount, 1);
+    assert.equal(history.length, 4);
+    assert.equal((await s.api(`${s.path}/retry`, {})).status, 409);
+    assert.equal((await s.api(s.path, {content: "unsafe edit"}, "PUT")).status, 409);
+    assert.equal((await s.api(s.path, null, "DELETE")).status, 409);
+    assert.equal((await s.api(`/chats/${s.chatId}/messages/${s.messages[3].id}`, {content: "Other answer edited"}, "PUT")).status, 200);
+    s.release();
+    while (true) { const result = await reader.read(); if (result.done) break; wire += new TextDecoder().decode(result.value); }
+    assert.match(wire, /event: done/);
+    assert.ok(wire.indexOf('"status":"searching"') < wire.indexOf("event: chunk"));
+    history = await s.history();
+    assert.deepEqual(history.map(m => m.id), s.messages.map(m => m.id));
+    assert.equal(history[1].content, "Partial answer");
+    assert.equal(history[1].versionCount, 2);
+    assert.equal(history[1].versionIndex, 1);
+    assert.equal(history[1].thinking, "New reasoning");
+    assert.equal(history[1].sources[0].url, "https://example.org/news/1");
+    assert.equal(s.searches, 1);
+    assert.ok(s.calls[0].messages.every(m => !m.content.includes("Later private question") && !m.content.includes("Original 2")));
+    assert.equal(s.calls[0].messages.at(-1).content, s.messages[0].content);
+    await s.api(`${s.path}/versions/0`, {});
+    history = await s.history();
+    assert.equal(history[1].content, s.bot.content);
+    assert.equal(history[1].thinking, "Original reasoning");
+    assert.equal(history[1].sources[0].url, "https://example.org/old");
+    const branch = await (await s.api(`${s.path}/branch`, {confirmed: true})).json();
+    assert.equal(branch.messages.at(-1).content, s.bot.content);
+    assert.equal(branch.messages.at(-1).thinking, "Original reasoning");
+    assert.deepEqual(branch.messages.at(-1).sources, history[1].sources);
+    await s.api(s.path, {content: "Edited original"}, "PUT");
+    await s.api(`${s.path}/versions/1`, {});
+    assert.equal((await s.history())[1].content, "Partial answer");
+    await s.api(`${s.path}/versions/0`, {});
+    assert.equal((await s.history())[1].content, "Edited original");
+    assert.equal((await s.api(`${s.path}/versions/99`, {})).status, 400);
+    s.setMode("success");
+    const third = await (await s.api(`${s.path}/retry/stream`, {})).text();
+    assert.match(third, /event: done/);
+    const thirdMessage = (await s.history())[1];
+    assert.equal(thirdMessage.versionCount, 3);
+    assert.equal(thirdMessage.versionIndex, 2);
+    assert.equal(thirdMessage.thinking, undefined);
+    assert.deepEqual(thirdMessage.sources, []);
+    const backup = (await (await s.api("/exports/workspace")).json()).workspace;
+    assert.equal((await s.api("/imports/workspace", {workspace: backup, allowDuplicates: true})).status, 200);
+    const copiedVersions = s.app.locals.db.prepare("SELECT * FROM chat_message_versions WHERE message_id != ?").all(s.bot.id);
+    assert.equal(copiedVersions.length, 3);
+    assert.equal(copiedVersions[0].thinking, "Original reasoning");
+    assert.equal(copiedVersions[1].thinking, "New reasoning");
+    const reopened = new Database(s.app.locals.db.serialize());
+    try {
+        runMigrations(reopened, {dropLegacyChats: false});
+        assert.equal(reopened.prepare("SELECT count(*) n FROM chat_message_versions WHERE message_id = ?").get(s.bot.id).n, 3);
+        assert.equal(reopened.prepare("SELECT retry_active_index FROM chat_messages WHERE id = ?").get(s.bot.id).retry_active_index, 2);
+    } finally { reopened.close(); }
+    await s.api(s.path, null, "DELETE");
+    assert.equal(s.app.locals.repositories.listMessageVersions(s.bot.id).length, 0);
+});
+
+test("failed, empty, stopped and disconnected regeneration preserve the saved answer and versions", async t => {
+    const s = await fixture(t);
+    const original = (await s.history())[1];
+    for (const mode of ["fail", "empty", "search-fail"]) {
+        s.setMode(mode);
+        const wire = await (await s.api(`${s.path}/retry/stream`, {reasoning: true, webSearch: mode === "search-fail"})).text();
+        assert.match(wire, /event: error/);
+        assert.doesNotMatch(wire, /event: done/);
+        assert.deepEqual((await s.history())[1], original);
+        assert.equal(s.app.locals.repositories.listMessageVersions(s.bot.id).length, 0);
+        if (mode === "search-fail") assert.match(wire, /"searchState":"failed"/);
+    }
+    const started = s.hold();
+    const response = await s.api(`${s.path}/retry/stream`, {});
+    await started;
+    assert.equal((await (await s.api(`/chats/${s.chatId}/stop`, {})).json()).stopped, true);
+    assert.match(await response.text(), /event: error/);
+    assert.deepEqual((await s.history())[1], original);
+    const disconnected = s.hold();
+    const disconnectResponse = await s.api(`${s.path}/retry/stream`, {});
+    await disconnected;
+    await disconnectResponse.body.cancel();
+    await new Promise(resolve => setTimeout(resolve, 20));
+    assert.deepEqual((await s.history())[1], original);
+    assert.equal(s.app.locals.repositories.listMessageVersions(s.bot.id).length, 0);
+});

test/regenerationUi.test.js

@@ -0,0 +1,95 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+import {readFile} from "node:fs/promises";
+import {regenerateInline} from "../public/app/regeneration.js";
+
+class Element {
+    constructor(tag = "div") {
+        this.tag = tag; this.children = []; this.style = {}; this.dataset = {}; this.attributes = {};
+        this.scrollTop = 0; this.bottom = 500; this.height = 200; this.top = 0;
+    }
+    get childNodes() { return this.children; }
+    get childElementCount() { return this.children.length; }
+    append(...children) { children.forEach(child => { child.parent = this; this.children.push(child); }); }
+    appendChild(child) { this.append(child); }
+    replaceChildren(...children) { this.children = []; this.append(...children); }
+    insertBefore(child, before) { child.parent = this; this.children.splice(this.children.indexOf(before), 0, child); }
+    remove() { this.parent.children = this.parent.children.filter(child => child !== this); }
+    cloneNode() { const clone = new Element(this.tag); clone.textContent = this.textContent; return clone; }
+    setAttribute(key, value) { this.attributes[key] = value; }
+    removeAttribute(key) { delete this.attributes[key]; }
+    getBoundingClientRect() { return {top: this.top, bottom: this.bottom, height: this.height}; }
+    querySelector(selector) {
+        for (const child of this.children) {
+            if (selector === child.tag || child.className?.split(" ").some(name => selector === `.${name}`)) return child;
+            const nested = child.querySelector(selector); if (nested) return nested;
+        }
+        return null;
+    }
+}
+
+test("inline renderer streams in place, follows only nearby output, commits or restores all original nodes", async t => {
+    const oldDocument = globalThis.document, oldFetch = globalThis.fetch;
+    t.after(() => { globalThis.document = oldDocument; globalThis.fetch = oldFetch; });
+    globalThis.document = {createElement: tag => new Element(tag)};
+    for (const outcome of ["success", "error", "disconnect"]) {
+        const scroller = new Element(), card = new Element(), later = new Element();
+        const header = new Element(), oldBody = new Element(), actions = new Element(), oldSources = new Element();
+        oldBody.textContent = "Old answer"; oldBody.className = "msg-content";
+        oldSources.className = "msg-sources";
+        card.append(header, oldBody, oldSources, actions);
+        scroller.append(card, later);
+        const original = [...card.childNodes];
+        let controller;
+        globalThis.fetch = async () => new Response(new ReadableStream({start(value) { controller = value; }}));
+        let completed;
+        const pending = regenerateInline({card, scroller, endpoint: "/retry/stream", payload: {},
+            renderContent: (body, content) => { body.textContent = content; card.bottom += 20; },
+            complete: message => { completed = message; card.replaceChildren(header, new Element(), actions); }});
+        const event = async (name, data) => {
+            controller.enqueue(new TextEncoder().encode(`event: ${name}\ndata: ${JSON.stringify(data)}\n\n`));
+            await new Promise(resolve => setTimeout(resolve, 0));
+        };
+        assert.equal(card.attributes["aria-busy"], "true");
+        assert.equal(card.querySelector(".msg-sources"), null);
+        assert.ok(!card.childNodes.includes(actions));
+        await event("status", {status: "searching"});
+        assert.equal(card.querySelector(".loading-shell").textContent, "Searching the web...");
+        await event("status", {status: "generating", sources: [{title: "New source", url: "https://example.org/new"}]});
+        assert.notEqual(card.querySelector(".msg-sources"), oldSources);
+        await event("thinking", {fullThinking: "New reasoning"});
+        assert.equal(card.querySelector(".msg-reasoning-content").textContent, "New reasoning");
+        await event("chunk", {fullReply: "Partial"});
+        assert.equal(card.querySelector(".msg-content").textContent, "Partial");
+        assert.equal(scroller.scrollTop, 20);
+        assert.equal(completed, undefined);
+        assert.deepEqual(scroller.childNodes, [card, later]);
+        card.bottom = 1000; // User has scrolled away from this response.
+        await event("chunk", {fullReply: "Partial answer"});
+        assert.equal(scroller.scrollTop, 20);
+        if (outcome === "success") {
+            await event("done", {reply: "Partial answer", message: {id: 2, content: "Partial answer"}});
+            await pending;
+            assert.equal(completed.content, "Partial answer");
+            assert.notDeepEqual(card.childNodes, original);
+        } else {
+            const rejection = assert.rejects(pending, outcome === "error" ? /Model failed/ : /interrupted/);
+            if (outcome === "error") await event("error", {error: "Model failed"});
+            else controller.close();
+            await rejection;
+            assert.deepEqual(card.childNodes, original);
+        }
+        assert.equal(card.attributes["aria-busy"], undefined);
+        assert.equal(card.style.minHeight, undefined);
+        assert.deepEqual(scroller.childNodes, [card, later]);
+    }
+});
+
+test("Regenerate uses inline activity and never reloads or appends a chat message", async () => {
+    const source = await readFile(new URL("../public/script.js", import.meta.url), "utf8");
+    const regenerate = source.slice(source.indexOf("async function retryLatestAssistantMessage"), source.indexOf("async function switchResponseVersion"));
+    assert.match(regenerate, /setLoadingState\(true, false\)/);
+    assert.match(regenerate, /regenerateInline/);
+    assert.match(regenerate, /detached: true/);
+    assert.doesNotMatch(regenerate, /setActiveChat\(|renderMessages\(|Generating another version/);
+});

test/webSearch.test.js

@@ -217,9 +217,10 @@ test("chat search integrates with streaming, persistence, personas, settings, re
     const botId = messages.at(-1).id;
     const retry = await api(`/chats/${chatId}/messages/${botId}/retry`, {});
     assert.equal(retry.status, 200);
-    assert.deepEqual(retry.body.message.sources, [source]);
+    assert.deepEqual(retry.body.message.sources, []);
+    await api(`/chats/${chatId}/messages/${botId}/versions/0`, {});
     assert.equal(searchCount, 1);
-    assert.ok(calls.at(-1).messages.some((message) => message.content.includes(source.snippet)));
+    assert.ok(!calls.at(-1).messages.some((message) => message.content.includes(source.snippet)));
     const branch = await api(`/chats/${chatId}/messages/${botId}/branch`, {confirmed: true});
     assert.deepEqual(branch.body.messages.at(-1).sources, [source]);
     const exported = (await api(`/exports/chats/${chatId}?format=json`)).body;
@@ -228,7 +229,7 @@ test("chat search integrates with streaming, persistence, personas, settings, re
     const imported = await api("/imports/workspace", {workspace, allowDuplicates: true});
     assert.equal(imported.status, 200);
     const allMessages = app.locals.db.prepare("SELECT sources FROM chat_messages WHERE sources IS NOT NULL").all();
-    assert.ok(allMessages.length >= 6);
+    assert.ok(allMessages.length >= 4);
     for (const message of allMessages) assert.deepEqual(JSON.parse(message.sources), [source]);
 
     emptySearch = true;