ai queue, better markdown parsing, refactor, better feedback
This commit is contained in:
parent
df49bc4157
commit
23ebd021f3
6 changed files with 273 additions and 173 deletions
|
|
@ -119,31 +119,17 @@ export const models: ModelInfo[] = [
|
|||
}
|
||||
];
|
||||
|
||||
const enSystemPrompt = `You are a plaintext-only, helpful assistant called {botName}.
|
||||
Current Date/Time (UTC): {date}
|
||||
|
||||
---
|
||||
|
||||
Respond to the user's message:
|
||||
{message}`
|
||||
|
||||
const ptSystemPrompt = `Você é um assistente de texto puro e útil chamado {botName}.
|
||||
Data/Hora atual (UTC): {date}
|
||||
|
||||
---
|
||||
|
||||
Responda à mensagem do usuário:
|
||||
{message}`
|
||||
|
||||
async function usingSystemPrompt(ctx: TextContext, db: NodePgDatabase<typeof schema>, botName: string): Promise<string> {
|
||||
async function usingSystemPrompt(ctx: TextContext, db: NodePgDatabase<typeof schema>, botName: string, message: string): Promise<string> {
|
||||
const user = await db.query.usersTable.findMany({ where: (fields, { eq }) => eq(fields.telegramId, String(ctx.from!.id)), limit: 1 });
|
||||
if (user.length === 0) await ensureUserInDb(ctx, db);
|
||||
const userData = user[0];
|
||||
const lang = userData?.languageCode || "en";
|
||||
const Strings = getStrings(lang);
|
||||
const utcDate = new Date().toISOString();
|
||||
const prompt = lang === "pt"
|
||||
? ptSystemPrompt.replace("{botName}", botName).replace("{date}", utcDate).replace("{message}", ctx.message.text)
|
||||
: enSystemPrompt.replace("{botName}", botName).replace("{date}", utcDate).replace("{message}", ctx.message.text);
|
||||
const prompt = Strings.ai.systemPrompt
|
||||
.replace("{botName}", botName)
|
||||
.replace("{date}", utcDate)
|
||||
.replace("{message}", message);
|
||||
return prompt;
|
||||
}
|
||||
|
||||
|
|
@ -156,6 +142,51 @@ export function sanitizeForJson(text: string): string {
|
|||
.replace(/\t/g, '\\t')
|
||||
}
|
||||
|
||||
function sanitizeMarkdownForTelegram(text: string): string {
|
||||
let sanitizedText = text;
|
||||
|
||||
const replacements: string[] = [];
|
||||
const addReplacement = (match: string): string => {
|
||||
replacements.push(match);
|
||||
return `___PLACEHOLDER_${replacements.length - 1}___`;
|
||||
};
|
||||
|
||||
sanitizedText = sanitizedText.replace(/```([\s\S]*?)```/g, addReplacement);
|
||||
sanitizedText = sanitizedText.replace(/`([^`]+)`/g, addReplacement);
|
||||
sanitizedText = sanitizedText.replace(/\[([^\]]+)\]\(([^)]+)\)/g, addReplacement);
|
||||
|
||||
const parts = sanitizedText.split(/(___PLACEHOLDER_\d+___)/g);
|
||||
const processedParts = parts.map(part => {
|
||||
if (part.match(/___PLACEHOLDER_\d+___/)) {
|
||||
return part;
|
||||
} else {
|
||||
let processedPart = part;
|
||||
processedPart = processedPart.replace(/^(#{1,6})\s+(.+)/gm, '*$2*');
|
||||
processedPart = processedPart.replace(/^(\s*)[-*]\s+/gm, '$1- ');
|
||||
processedPart = processedPart.replace(/\*\*(.*?)\*\*/g, '*$1*');
|
||||
processedPart = processedPart.replace(/__(.*?)__/g, '*$1*');
|
||||
processedPart = processedPart.replace(/(^|\s)\*(?!\*)([^*]+?)\*(?!\*)/g, '$1_$2_');
|
||||
processedPart = processedPart.replace(/(^|\s)_(?!_)([^_]+?)_(?!_)/g, '$1_$2_');
|
||||
processedPart = processedPart.replace(/~~(.*?)~~/g, '~$1~');
|
||||
processedPart = processedPart.replace(/^\s*┃/gm, '>');
|
||||
processedPart = processedPart.replace(/^>\s?/gm, '> ');
|
||||
|
||||
return processedPart;
|
||||
}
|
||||
});
|
||||
|
||||
sanitizedText = processedParts.join('');
|
||||
|
||||
sanitizedText = sanitizedText.replace(/___PLACEHOLDER_(\d+)___/g, (_, idx) => replacements[Number(idx)]);
|
||||
|
||||
const codeBlockCount = (sanitizedText.match(/```/g) || []).length;
|
||||
if (codeBlockCount % 2 !== 0) {
|
||||
sanitizedText += '\n```';
|
||||
}
|
||||
|
||||
return sanitizedText;
|
||||
}
|
||||
|
||||
export async function preChecks() {
|
||||
const envs = [
|
||||
"ollamaApi",
|
||||
|
|
@ -232,7 +263,7 @@ function extractAxiosErrorMessage(error: unknown): string {
|
|||
}
|
||||
|
||||
function escapeMarkdown(text: string): string {
|
||||
return text.replace(/([*_])/g, '\\$1');
|
||||
return text.replace(/([_*\[\]()`>#\+\-=|{}.!~])/g, '\\$1');
|
||||
}
|
||||
|
||||
function containsUrls(text: string): boolean {
|
||||
|
|
@ -244,10 +275,14 @@ async function getResponse(prompt: string, ctx: TextContext, replyGenerating: Me
|
|||
if (!ctx.chat) {
|
||||
return {
|
||||
success: false,
|
||||
error: Strings.unexpectedErr.replace("{error}", "No chat found"),
|
||||
error: Strings.unexpectedErr.replace("{error}", Strings.ai.noChatFound),
|
||||
};
|
||||
}
|
||||
const modelHeader = `🤖 *${model}* | 🌡️ *${aiTemperature}*\n\n`;
|
||||
let status = Strings.ai.statusWaitingRender;
|
||||
let modelHeader = Strings.ai.modelHeader
|
||||
.replace("{model}", model)
|
||||
.replace("{temperature}", aiTemperature)
|
||||
.replace("{status}", status) + "\n\n";
|
||||
const urlWarning = containsUrls(originalMessage) ? Strings.ai.urlWarning : '';
|
||||
|
||||
try {
|
||||
|
|
@ -267,8 +302,9 @@ async function getResponse(prompt: string, ctx: TextContext, replyGenerating: Me
|
|||
);
|
||||
let fullResponse = "";
|
||||
let thoughts = "";
|
||||
let lastUpdate = Date.now();
|
||||
let lastUpdateCharCount = 0;
|
||||
let sentHeader = false;
|
||||
let firstChunk = true;
|
||||
const stream: NodeJS.ReadableStream = aiResponse.data as any;
|
||||
for await (const chunk of stream) {
|
||||
const lines = chunk.toString().split('\n');
|
||||
|
|
@ -293,7 +329,6 @@ async function getResponse(prompt: string, ctx: TextContext, replyGenerating: Me
|
|||
logger.logThinking(ctx.chat.id, replyGenerating.message_id, false);
|
||||
}
|
||||
}
|
||||
const now = Date.now();
|
||||
if (ln.response) {
|
||||
if (model === thinking_model) {
|
||||
let patchedThoughts = ln.response;
|
||||
|
|
@ -306,20 +341,51 @@ async function getResponse(prompt: string, ctx: TextContext, replyGenerating: Me
|
|||
} else {
|
||||
fullResponse += ln.response;
|
||||
}
|
||||
if (now - lastUpdate >= 5000 || !sentHeader) {
|
||||
if (firstChunk) {
|
||||
status = Strings.ai.statusWaitingRender;
|
||||
modelHeader = Strings.ai.modelHeader
|
||||
.replace("{model}", model)
|
||||
.replace("{temperature}", aiTemperature)
|
||||
.replace("{status}", status) + "\n\n";
|
||||
await rateLimiter.editMessageWithRetry(
|
||||
ctx,
|
||||
ctx.chat.id,
|
||||
replyGenerating.message_id,
|
||||
modelHeader + urlWarning + fullResponse,
|
||||
modelHeader + urlWarning + escapeMarkdown(fullResponse),
|
||||
{ parse_mode: 'Markdown' }
|
||||
);
|
||||
lastUpdate = now;
|
||||
lastUpdateCharCount = fullResponse.length;
|
||||
sentHeader = true;
|
||||
firstChunk = false;
|
||||
continue;
|
||||
}
|
||||
const updateEveryChars = Number(process.env.updateEveryChars) || 100;
|
||||
if (fullResponse.length - lastUpdateCharCount >= updateEveryChars || !sentHeader) {
|
||||
await rateLimiter.editMessageWithRetry(
|
||||
ctx,
|
||||
ctx.chat.id,
|
||||
replyGenerating.message_id,
|
||||
modelHeader + urlWarning + escapeMarkdown(fullResponse),
|
||||
{ parse_mode: 'Markdown' }
|
||||
);
|
||||
lastUpdateCharCount = fullResponse.length;
|
||||
sentHeader = true;
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
status = Strings.ai.statusRendering;
|
||||
modelHeader = Strings.ai.modelHeader
|
||||
.replace("{model}", model)
|
||||
.replace("{temperature}", aiTemperature)
|
||||
.replace("{status}", status) + "\n\n";
|
||||
await rateLimiter.editMessageWithRetry(
|
||||
ctx,
|
||||
ctx.chat.id,
|
||||
replyGenerating.message_id,
|
||||
modelHeader + urlWarning + escapeMarkdown(fullResponse),
|
||||
{ parse_mode: 'Markdown' }
|
||||
);
|
||||
return {
|
||||
success: true,
|
||||
response: fullResponse,
|
||||
|
|
@ -360,7 +426,7 @@ async function getResponse(prompt: string, ctx: TextContext, replyGenerating: Me
|
|||
console.log(`[✨ AI] ${model} pulled successfully`);
|
||||
return {
|
||||
success: true,
|
||||
response: `✅ Pulled ${escapeMarkdown(model)} successfully, please retry the command.`,
|
||||
response: Strings.ai.pulled.replace("{model}", escapeMarkdown(model)),
|
||||
};
|
||||
}
|
||||
}
|
||||
|
|
@ -376,16 +442,18 @@ async function handleAiReply(ctx: TextContext, model: string, prompt: string, re
|
|||
const aiResponse = await getResponse(prompt, ctx, replyGenerating, model, aiTemperature, originalMessage);
|
||||
if (!aiResponse) return;
|
||||
if (!ctx.chat) return;
|
||||
const modelHeader = `🤖 *${model}* | 🌡️ *${aiTemperature}*\n\n`;
|
||||
|
||||
const urlWarning = containsUrls(originalMessage) ? Strings.ai.urlWarning : '';
|
||||
|
||||
if (aiResponse.success && aiResponse.response) {
|
||||
const status = Strings.ai.statusComplete;
|
||||
const modelHeader = Strings.ai.modelHeader
|
||||
.replace("{model}", model)
|
||||
.replace("{temperature}", aiTemperature)
|
||||
.replace("{status}", status) + "\n\n";
|
||||
const urlWarning = containsUrls(originalMessage) ? Strings.ai.urlWarning : '';
|
||||
await rateLimiter.editMessageWithRetry(
|
||||
ctx,
|
||||
ctx.chat.id,
|
||||
replyGenerating.message_id,
|
||||
modelHeader + urlWarning + aiResponse.response,
|
||||
modelHeader + urlWarning + sanitizeMarkdownForTelegram(aiResponse.response),
|
||||
{ parse_mode: 'Markdown' }
|
||||
);
|
||||
return;
|
||||
|
|
@ -425,109 +493,112 @@ export function getModelLabelByName(name: string): string {
|
|||
export default (bot: Telegraf<Context>, db: NodePgDatabase<typeof schema>) => {
|
||||
const botName = bot.botInfo?.first_name && bot.botInfo?.last_name ? `${bot.botInfo.first_name} ${bot.botInfo.last_name}` : "Kowalski"
|
||||
|
||||
bot.command(["ask", "think"], spamwatchMiddleware, async (ctx) => {
|
||||
if (!ctx.message || !('text' in ctx.message)) return
|
||||
const isAsk = ctx.message.text.startsWith("/ask")
|
||||
const model = isAsk ? flash_model : thinking_model
|
||||
const textCtx = ctx as TextContext
|
||||
const reply_to_message_id = replyToMessageId(textCtx)
|
||||
const { user, Strings, aiTemperature } = await getUserWithStringsAndModel(textCtx, db)
|
||||
const message = textCtx.message.text
|
||||
const author = ("@" + ctx.from?.username) || ctx.from?.first_name
|
||||
interface AiRequest {
|
||||
task: () => Promise<void>;
|
||||
ctx: TextContext;
|
||||
wasQueued: boolean;
|
||||
}
|
||||
|
||||
logger.logCmdStart(author, model === flash_model ? "ask" : "think")
|
||||
const requestQueue: AiRequest[] = [];
|
||||
let isProcessing = false;
|
||||
|
||||
async function processQueue() {
|
||||
if (isProcessing || requestQueue.length === 0) {
|
||||
return;
|
||||
}
|
||||
|
||||
isProcessing = true;
|
||||
const { task, ctx, wasQueued } = requestQueue.shift()!;
|
||||
const { Strings } = await getUserWithStringsAndModel(ctx, db);
|
||||
const reply_to_message_id = replyToMessageId(ctx);
|
||||
|
||||
try {
|
||||
if (wasQueued) {
|
||||
await ctx.reply(Strings.ai.startingProcessing, {
|
||||
...(reply_to_message_id && { reply_parameters: { message_id: reply_to_message_id } }),
|
||||
parse_mode: 'Markdown'
|
||||
});
|
||||
}
|
||||
await task();
|
||||
} catch (error) {
|
||||
console.error("[✨ AI | !] Error processing task:", error);
|
||||
const errorMessage = error instanceof Error ? error.message : String(error);
|
||||
await ctx.reply(Strings.unexpectedErr.replace("{error}", errorMessage), {
|
||||
...(reply_to_message_id && { reply_parameters: { message_id: reply_to_message_id } }),
|
||||
parse_mode: 'Markdown'
|
||||
});
|
||||
} finally {
|
||||
isProcessing = false;
|
||||
processQueue();
|
||||
}
|
||||
}
|
||||
|
||||
async function aiCommandHandler(ctx: TextContext, command: 'ask' | 'think' | 'ai') {
|
||||
const reply_to_message_id = replyToMessageId(ctx);
|
||||
const { user, Strings, customAiModel, aiTemperature } = await getUserWithStringsAndModel(ctx, db);
|
||||
const message = ctx.message.text;
|
||||
const author = ("@" + ctx.from?.username) || ctx.from?.first_name || "Unknown";
|
||||
|
||||
let model: string;
|
||||
let fixedMsg: string;
|
||||
|
||||
if (command === 'ai') {
|
||||
model = customAiModel || flash_model;
|
||||
fixedMsg = message.replace(/^\/ai(@\w+)?\s*/, "").trim();
|
||||
logger.logCmdStart(author, "ask");
|
||||
} else {
|
||||
model = command === 'ask' ? flash_model : thinking_model;
|
||||
fixedMsg = message.replace(/^\/(ask|think)(@\w+)?\s*/, "").trim();
|
||||
logger.logCmdStart(author, command);
|
||||
}
|
||||
|
||||
if (!process.env.ollamaApi) {
|
||||
await ctx.reply(Strings.ai.disabled, {
|
||||
parse_mode: 'Markdown',
|
||||
...({ reply_to_message_id })
|
||||
})
|
||||
return
|
||||
await ctx.reply(Strings.ai.disabled, { parse_mode: 'Markdown', ...(reply_to_message_id && { reply_parameters: { message_id: reply_to_message_id } }) });
|
||||
return;
|
||||
}
|
||||
|
||||
if (!user.aiEnabled) {
|
||||
await ctx.reply(Strings.ai.disabledForUser, {
|
||||
parse_mode: 'Markdown',
|
||||
...({ reply_to_message_id })
|
||||
})
|
||||
return
|
||||
await ctx.reply(Strings.ai.disabledForUser, { parse_mode: 'Markdown', ...(reply_to_message_id && { reply_parameters: { message_id: reply_to_message_id } }) });
|
||||
return;
|
||||
}
|
||||
|
||||
const fixedMsg = message.replace(/^\/(ask|think)(@\w+)?\s*/, "").trim()
|
||||
if (fixedMsg.length < 1) {
|
||||
await ctx.reply(Strings.ai.askNoMessage, {
|
||||
parse_mode: 'Markdown',
|
||||
...({ reply_to_message_id })
|
||||
})
|
||||
return
|
||||
await ctx.reply(Strings.ai.askNoMessage, { parse_mode: 'Markdown', ...(reply_to_message_id && { reply_parameters: { message_id: reply_to_message_id } }) });
|
||||
return;
|
||||
}
|
||||
|
||||
const replyGenerating = await ctx.reply(Strings.ai.askGenerating.replace("{model}", model), {
|
||||
parse_mode: 'Markdown',
|
||||
...({ reply_to_message_id })
|
||||
})
|
||||
|
||||
logger.logPrompt(fixedMsg)
|
||||
|
||||
const prompt = sanitizeForJson(await usingSystemPrompt(textCtx, db, botName))
|
||||
await handleAiReply(textCtx, model, prompt, replyGenerating, aiTemperature, fixedMsg)
|
||||
})
|
||||
|
||||
bot.command(["ai"], spamwatchMiddleware, async (ctx) => {
|
||||
try {
|
||||
if (!ctx.message || !("text" in ctx.message)) return
|
||||
const textCtx = ctx as TextContext
|
||||
const reply_to_message_id = replyToMessageId(textCtx)
|
||||
const { user, Strings, customAiModel, aiTemperature } = await getUserWithStringsAndModel(textCtx, db)
|
||||
const message = textCtx.message.text
|
||||
const author = ("@" + ctx.from?.username) || ctx.from?.first_name
|
||||
|
||||
logger.logCmdStart(author, "ask")
|
||||
|
||||
if (!process.env.ollamaApi) {
|
||||
await ctx.reply(Strings.ai.disabled, {
|
||||
parse_mode: 'Markdown',
|
||||
...({ reply_to_message_id })
|
||||
})
|
||||
return
|
||||
}
|
||||
|
||||
if (!user.aiEnabled) {
|
||||
await ctx.reply(Strings.ai.disabledForUser, {
|
||||
parse_mode: 'Markdown',
|
||||
...({ reply_to_message_id })
|
||||
})
|
||||
return
|
||||
}
|
||||
|
||||
const fixedMsg = message.replace(/^\/ai(@\w+)?\s*/, "").trim()
|
||||
if (fixedMsg.length < 1) {
|
||||
await ctx.reply(Strings.ai.askNoMessage, {
|
||||
parse_mode: 'Markdown',
|
||||
...({ reply_to_message_id })
|
||||
})
|
||||
return
|
||||
}
|
||||
|
||||
const modelLabel = getModelLabelByName(customAiModel)
|
||||
const task = async () => {
|
||||
const modelLabel = getModelLabelByName(model);
|
||||
const replyGenerating = await ctx.reply(Strings.ai.askGenerating.replace("{model}", modelLabel), {
|
||||
parse_mode: 'Markdown',
|
||||
...({ reply_to_message_id })
|
||||
})
|
||||
...(reply_to_message_id && { reply_parameters: { message_id: reply_to_message_id } })
|
||||
});
|
||||
logger.logPrompt(fixedMsg);
|
||||
const prompt = sanitizeForJson(await usingSystemPrompt(ctx, db, botName, fixedMsg));
|
||||
await handleAiReply(ctx, model, prompt, replyGenerating, aiTemperature, fixedMsg);
|
||||
};
|
||||
|
||||
logger.logPrompt(fixedMsg)
|
||||
|
||||
const prompt = sanitizeForJson(await usingSystemPrompt(textCtx, db, botName))
|
||||
await handleAiReply(textCtx, customAiModel, prompt, replyGenerating, aiTemperature, fixedMsg)
|
||||
} catch (err) {
|
||||
const Strings = getStrings(languageCode(ctx));
|
||||
if (ctx && ctx.reply) {
|
||||
try {
|
||||
await ctx.reply(Strings.unexpectedErr.replace("{error}", (err && err.message ? err.message : String(err))), { parse_mode: 'Markdown' })
|
||||
} catch (e) {
|
||||
console.error("[✨ AI | !] Failed to send error reply:", e)
|
||||
}
|
||||
}
|
||||
if (isProcessing) {
|
||||
requestQueue.push({ task, ctx, wasQueued: true });
|
||||
const position = requestQueue.length;
|
||||
await ctx.reply(Strings.ai.inQueue.replace("{position}", String(position)), {
|
||||
parse_mode: 'Markdown',
|
||||
...(reply_to_message_id && { reply_parameters: { message_id: reply_to_message_id } })
|
||||
});
|
||||
} else {
|
||||
requestQueue.push({ task, ctx, wasQueued: false });
|
||||
processQueue();
|
||||
}
|
||||
})
|
||||
}
|
||||
|
||||
bot.command(["ask", "think"], spamwatchMiddleware, async (ctx) => {
|
||||
if (!ctx.message || !('text' in ctx.message)) return;
|
||||
const command = ctx.message.text.startsWith('/ask') ? 'ask' : 'think';
|
||||
await aiCommandHandler(ctx as TextContext, command);
|
||||
});
|
||||
|
||||
bot.command(["ai"], spamwatchMiddleware, async (ctx) => {
|
||||
if (!ctx.message || !('text' in ctx.message)) return;
|
||||
await aiCommandHandler(ctx as TextContext, 'ai');
|
||||
});
|
||||
}
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue