Fretux/KrisHD_AI
3
0
Fork 0
Code Issues 3 Pull Requests Actions Packages Projects 1 Releases Wiki Activity

show reasoning in frontend

Browse Source
This commit is contained in:
fretuks
2026-09-21 19:50:16 +02:00
parent 8c13f50bd1
commit ceb3c752ff
12 changed files with 104 additions and 40 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/services/chatMessageHelpers.js
+1
View File
@@ -41,6 +41,7 @@ export function formatChatMessage(row) {
id: row.id,
role: row.role,
content: row.content,
...(row.thinking ? {thinking: row.thinking} : {}),
...(row.sources ? {sources: parseSources(row.sources)} : {}),
modelName: row.model_name || null,
retryVariants,
src/services/chatService.js
+31 -17
View File
@@ -168,7 +168,7 @@ 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}) => {
const generateRoleplayOpener = async ({selectedModel = "qwen3:14b", assistantPersona, userPersona, scenarioPrompt, sceneSummary, onThinking = null}) => {
const baseMessages = [
{role: "system", content: buildPersonaPrompt(assistantPersona)},
...(userPersona ? [{role: "system", content: buildUserPersonaPrompt(userPersona)}] : []),
@@ -176,11 +176,12 @@ export function createChatService(repositories, modelService, config, webSearchP
{role: "user", content: buildRoleplayOpenerPrompt({assistantPersona, userPersona, scenarioPrompt, sceneSummary})}
];
let opener = await modelService.generateReply(selectedModel, baseMessages);
let opener = await modelService.generateReply(selectedModel, baseMessages, {onThinking});
if (!containsUserVoiceInRoleplayOpener(opener, userPersona)) {
return opener;
}
onThinking?.("", "");
return modelService.generateReply(selectedModel, [
...baseMessages,
{role: "assistant", content: opener || ""},
@@ -188,15 +189,16 @@ export function createChatService(repositories, modelService, config, webSearchP
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 = {}}) => {
let reply = await modelService.generateReply(selectedModel, messagesPayload, {signal, generation});
const generateValidatedRoleplayReply = async ({selectedModel, messagesPayload, userPersona, signal = null, generation = {}, onThinking = null}) => {
let reply = await modelService.generateReply(selectedModel, messagesPayload, {signal, generation, onThinking});
if (reply && !containsUserVoiceInRoleplayOpener(reply, userPersona)) {
return reply;
}
onThinking?.("", "");
reply = await modelService.generateReply(selectedModel, [
...messagesPayload,
{role: "assistant", content: reply || ""},
@@ -204,7 +206,7 @@ export function createChatService(repositories, modelService, config, webSearchP
role: "user",
content: "Rewrite your last reply. Hard rule: only the assistant character may speak or act. Never write or imply user speech, thoughts, feelings, reactions, decisions, or actions. Stop before the user's reply."
}
], {signal, generation});
], {signal, generation, onThinking});
if (!reply || containsUserVoiceInRoleplayOpener(reply, userPersona)) {
return null;
@@ -283,7 +285,9 @@ 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; },
selectedModel,
assistantPersona: activePersona,
userPersona: activeUserPersona,
@@ -293,7 +297,7 @@ export function createChatService(repositories, modelService, config, webSearchP
if (!opener || containsUserVoiceInRoleplayOpener(opener, activeUserPersona)) {
return {error: "Failed to regenerate message"};
}
return {fullReply: opener, promptMessageId: null};
return {fullReply: opener, thinking, promptMessageId: null};
}
const {messagesPayload} = buildConversationContext({
@@ -306,15 +310,17 @@ export function createChatService(repositories, modelService, config, webSearchP
});
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})
: await modelService.generateReply(selectedModel, styledPayload, {signal, generation});
? await generateValidatedRoleplayReply({selectedModel, messagesPayload: styledPayload, userPersona: activeUserPersona, signal, generation, onThinking})
: await modelService.generateReply(selectedModel, styledPayload, {signal, generation, onThinking});
if (!fullReply) return {error: "Failed to regenerate message"};
return {fullReply, promptMessageId: promptMessage.id};
return {fullReply, thinking, promptMessageId: promptMessage.id};
};
const generateChatTurn = ({user, chatId, message, model, webSearch = false, onStatus = null, onChunk = null, signal = null, parentMessageId = undefined}) =>
const generateChatTurn = ({user, chatId, message, model, webSearch = false, onStatus = null, onChunk = null, onThinking = null, signal = null, parentMessageId = undefined}) =>
withChatGenerationLock(chatId, async () => {
await (summaryQueues.get(chatId) || Promise.resolve());
const session = repositories.getChat(chatId, user);
@@ -335,6 +341,11 @@ export function createChatService(repositories, modelService, config, webSearchP
const userMessageId = Number(inserted?.lastInsertRowid);
try {
let sources = null;
let thinking = "";
const captureThinking = (chunk, fullThinking) => {
thinking = fullThinking;
onThinking?.(chunk, fullThinking);
};
let payload;
const fullReply = await withActiveGeneration(chatId, signal, async (activeSignal) => {
if (webSearch) {
@@ -375,11 +386,12 @@ export function createChatService(repositories, modelService, config, webSearchP
messagesPayload: payload.messagesPayload,
userPersona: payload.activeUserPersona,
signal: activeSignal,
generation
generation,
onThinking: captureThinking
})
: onChunk
? modelService.streamReply(effectiveModel, payload.messagesPayload, onChunk, {signal: activeSignal, generation})
: modelService.generateReply(effectiveModel, payload.messagesPayload, {signal: activeSignal, generation});
? modelService.streamReply(effectiveModel, payload.messagesPayload, onChunk, {signal: activeSignal, generation, onThinking: captureThinking})
: modelService.generateReply(effectiveModel, payload.messagesPayload, {signal: activeSignal, generation, onThinking: captureThinking});
});
if (!fullReply) {
@@ -394,7 +406,7 @@ export function createChatService(repositories, modelService, config, webSearchP
retryRetriesUsed: 0,
retryPromptMessageId: null
};
repositories.completeChatTurn(chatId, userMessageId, fullReply, retryState, effectiveModel, sources);
repositories.completeChatTurn(chatId, userMessageId, fullReply, retryState, effectiveModel, sources, thinking);
repositories.touchChat(chatId, user);
scheduleContextSummaryUpdate({user, chatId, selectedModel: effectiveModel});
return {reply: fullReply, userMessageId, ...(sources ? {sources} : {})};
@@ -471,6 +483,7 @@ export function createChatService(repositories, modelService, config, webSearchP
);
parentMessageId = Number(inserted.lastInsertRowid);
if (sourceMessage.sources) repositories.updateChatMessageSources(clonedChat.id, parentMessageId, parseSources(sourceMessage.sources));
if (sourceMessage.thinking) repositories.updateChatMessageThinking(clonedChat.id, parentMessageId, sourceMessage.thinking);
}
} catch (error) {
repositories.deleteChat(clonedChat.id, user);
@@ -498,8 +511,8 @@ export function createChatService(repositories, modelService, config, webSearchP
async sendChatMessage({user, chatId, message, model, webSearch = false}) {
return generateChatTurn({user, chatId, message, model, webSearch});
},
async streamChatMessage({user, chatId, message, model, webSearch = false, onStatus, onChunk, signal = null}) {
return generateChatTurn({user, chatId, message, model, webSearch, onStatus, onChunk, signal});
async streamChatMessage({user, chatId, message, model, webSearch = false, onStatus, onChunk, onThinking, signal = null}) {
return generateChatTurn({user, chatId, message, model, webSearch, onStatus, onChunk, onThinking, signal});
},
async retryMessage({user, chatId, selectedModel, targetMessage, retryStyle = null}) {
return withChatGenerationLock(chatId, async () => {
@@ -532,6 +545,7 @@ export function createChatService(repositories, modelService, config, webSearchP
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)));
src/services/modelService.js
+17 -13
View File
@@ -74,7 +74,7 @@ export function createModelService(config, overrides = {}) {
}
};
const readStreamingReply = async (response, onChunk = null) => {
const readStreamingReply = async (response, onChunk = null, onThinking = null) => {
if (!response.body) {
throw new ModelServiceError("MODEL_BAD_RESPONSE", "Model backend returned no body.", 502);
}
@@ -83,6 +83,18 @@ export function createModelService(config, overrides = {}) {
const decoder = new TextDecoder();
let buffer = "";
let fullReply = "";
let fullThinking = "";
const consume = (line) => {
const data = JSON.parse(line);
if (data.message?.thinking) {
fullThinking += data.message.thinking;
onThinking?.(data.message.thinking, fullThinking);
}
if (data.message?.content) {
fullReply += data.message.content;
onChunk?.(data.message.content, fullReply);
}
};
while (true) {
const {value, done} = await reader.read();
@@ -93,20 +105,12 @@ export function createModelService(config, overrides = {}) {
for (const line of lines) {
const trimmed = line.trim();
if (!trimmed) continue;
const data = JSON.parse(trimmed);
if (data.message?.content) {
fullReply += data.message.content;
if (onChunk) onChunk(data.message.content, fullReply);
}
consume(trimmed);
}
}
if (buffer.trim()) {
const data = JSON.parse(buffer.trim());
if (data.message?.content) {
fullReply += data.message.content;
if (onChunk) onChunk(data.message.content, fullReply);
}
consume(buffer.trim());
}
return fullReply.trim();
@@ -146,7 +150,7 @@ export function createModelService(config, overrides = {}) {
throw new ModelServiceError("MODEL_UPSTREAM_ERROR", `Model backend returned ${response.status}.`, 502);
}
return readStreamingReply(response);
return readStreamingReply(response, null, options.onThinking);
}, options.signal);
lastHealth = {ok: true, checkedAt: new Date().toISOString(), error: null, latencyMs: Date.now() - startedAt};
return fullReply;
@@ -184,7 +188,7 @@ export function createModelService(config, overrides = {}) {
throw new ModelServiceError("MODEL_UPSTREAM_ERROR", `Model backend returned ${response.status}.`, 502);
}
return readStreamingReply(response, onChunk);
return readStreamingReply(response, onChunk, options.onThinking);
}, options.signal);
lastHealth = {ok: true, checkedAt: new Date().toISOString(), error: null, latencyMs: Date.now() - startedAt};
return fullReply;
src/services/workspaceImportService.js
+1
View File
@@ -85,6 +85,7 @@ function validateMessage(message, chatIndex, messageIndex, errors) {
role: role === "assistant" ? "bot" : role,
content,
model_name: optionalText(message.model_name || message.modelName),
thinking: optionalText(message.thinking),
...(message.sources ? {sources: parseSources(message.sources)} : {}),
retryVariants,
retryActiveIndex: Number.isInteger(Number(message.retryActiveIndex ?? message.retry_active_index))