melod1n/tg-chat-bot
1
0
Fork 0
Code Issues Actions Packages Projects Releases Wiki Activity

feat: add Ollama audio transcription and runtime config reload

Browse Source 
- add audio capability reporting for Ollama models
- support Telegram voice messages via ffmpeg conversion and Ollama transcription
- add USE_SYSTEM_PROMPT toggle and runtime reloading of .env/system prompt settings
- support ollama_options.json for custom Ollama request options
- improve Telegram MarkdownV2 escaping and formatting preservation
- add environment setters for AI provider credentials and models
- show audio capability in info/model commands
This commit is contained in:
Danil Nikolaev
2026-05-01 05:09:10 +03:00
parent 382e00ce31
commit 0f91e43ea0
11 changed files with 937 additions and 217 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/commands/ollama-chat.ts
+54 -22
View File
@@ -14,6 +14,10 @@ import {Cancel} from "../callback_commands/cancel";
import {OllamaCancel} from "../callback_commands/ollama-cancel";
import {OllamaGetModel} from "./ollama-get-model";
import {ChatCommand} from "../base/chat-command";
import {MessagePart} from "../common/message-part";
import {Options} from "ollama";
import fs from "node:fs";
import path from "node:path";
export class OllamaChat extends ChatCommand {
command = ["ollamaThink", "ollama"];
@@ -27,25 +31,38 @@ export class OllamaChat extends ChatCommand {
return this.executeOllama(msg, match?.[3], match?.[1]?.toLowerCase()?.startsWith("ollamathink"));
}
async executeOllama(msg: Message, text: string, think: boolean = false): Promise<void> {
if (!text || text.trim().length === 0) return;
async executeOllama(msg: Message, text: string, think: boolean = false, voiceB64?: string): Promise<void> {
if ((!text || text.trim().length === 0) && !voiceB64) return;
const chatId = msg.chat.id;
const storedMsg = await MessageStore.get(chatId, msg.message_id);
const messageParts = await collectReplyChainText(storedMsg);
console.log("MESSAGE PARTS", messageParts);
const chatMessages = messageParts.map(part => {
let messageParts: MessagePart[] = [];
if (!voiceB64) {
messageParts = await collectReplyChainText(storedMsg);
console.log("MESSAGE PARTS", messageParts);
}
const chatMessages = !voiceB64 ? messageParts.map(part => {
return {
role: part.bot ? "assistant" : "user",
content: (Environment.USE_NAMES_IN_PROMPT && !part.bot ? `MESSAGE FROM USER "${part.name}":\n` : "") + part.content,
content: (Environment.USE_NAMES_IN_PROMPT && !part.bot ? `"${part.name}":\n` : "") + part.content,
images: part.images
};
});
}) : [
{
role: "user",
content: "Transcribe this audio file",
images: [voiceB64]
}
];
chatMessages.reverse();
if (Environment.SYSTEM_PROMPT) {
console.log("PARTS", chatMessages);
if (Environment.SYSTEM_PROMPT && !voiceB64 && Environment.USE_SYSTEM_PROMPT) {
chatMessages.unshift({role: "system", content: Environment.SYSTEM_PROMPT, images: []});
}
@@ -97,16 +114,28 @@ export class OllamaChat extends ChatCommand {
waitMessage = await replyToMessage({
message: msg,
text: (!think && imagesCount) ?
text: (!think && imagesCount && !voiceB64) ?
imagesCount > 1 ? Environment.analyzingPicturesText : Environment.analyzingPictureText
: Environment.waitThinkText
: voiceB64 ? Environment.transcribingAudioText : Environment.waitThinkText
});
let options: Partial<Options> | null = null
try {
const optionsPath = path.join(Environment.DATA_PATH, "ollama_options.json");
if (fs.existsSync(optionsPath)) {
options = JSON.parse(fs.readFileSync(optionsPath).toString());
}
} catch (e) {
logError(e);
}
const stream = await ollama.chat({
model: think ? Environment.OLLAMA_THINK_MODEL : imagesCount ? Environment.OLLAMA_IMAGE_MODEL : Environment.OLLAMA_MODEL,
stream: true,
think: think,
messages: chatMessages,
options: options
});
const newRequest = {
@@ -171,7 +200,7 @@ export class OllamaChat extends ChatCommand {
chat_id: chatId,
message_id: waitMessage.message_id,
text: "🤔 Размышляю...",
parse_mode: "Markdown",
parse_mode: "MarkdownV2",
reply_markup: cancelMarkup
}).catch(logError);
}
@@ -210,29 +239,32 @@ export class OllamaChat extends ChatCommand {
console.log("ended", true);
}
const diff = Math.abs(Date.now() - startTime) / 1000;
await editor.tick();
await editor.stop();
console.log(`aborted request ${uuid}:`, abortOllamaRequest(uuid));
waitMessage.reply_to_message = msg;
waitMessage.text = currentText;
await MessageStore.put(waitMessage);
if (Environment.SEND_TIME_TOOK) {
await replyToMessage({message: waitMessage, text: `⏱️ ${diff}s`});
}
break;
}
}
} finally {
const diff = Math.abs(Date.now() - startTime) / 1000;
await editor.tick();
await editor.stop();
waitMessage.reply_to_message = msg;
waitMessage.text = currentText;
await MessageStore.put(waitMessage);
if (Environment.SEND_TIME_TOOK) {
await replyToMessage({message: waitMessage, text: `⏱️ ${diff}s`});
}
await bot.editMessageReplyMarkup({
chat_id: chatId,
message_id: waitMessage.message_id,
reply_markup: {inline_keyboard: []}
}).catch(logError);
console.log(`aborted request ${uuid}:`, abortOllamaRequest(uuid));
}
} catch (error) {
if (error.message.toLowerCase().includes("aborted")) return;