An always-on second brain you talk to. Voice notes in Telegram → typed, linked knowledge in your Obsidian vault. Runs 24/7 on the Claude subscription you already have.
Python ai-agentsai-automationclaudeclaude-codeebbinghausknowledge-graphllmmemory-systemobsidianpersonal-assistantpkmproductivitypythonsecond-braintelegram-botvoice-assistant
feat(chat): media input — photos, files, forwards; never silent
smixseeb28ffb54cf
diff --git a/README.md b/README.mdindex 6555585..06b43d8 100644--- a/README.md+++ b/README.md@@ -24,7 +24,7 @@ The real issue: organizing takes more effort than thinking. So the thinking neve## What this actually does-You talk to a Telegram bot. Voice or text - whatever is natural. You don't think about categories, tags, or where things go.+You talk to a Telegram bot. Voice, text, photos, files, forwarded posts - whatever is natural. You don't think about categories, tags, or where things go.The agent handles everything else:@@ -93,6 +93,7 @@ You don't run maintenance. The agent does.|----------|-----------|| Voice note about a client call | Transcribes, creates CRM card with a follow-up || Quick text: "idea for the Q2 campaign" | Saves to ideas folder, links to related notes |+| Forwarded post or a photo of a whiteboard | Saves the file to the vault, the agent reads it and files the takeaways || "Remind me about the call tomorrow at 3pm" | Creates a one-shot scheduled job, pings you in Telegram || "Every weekday at 18:30 check my inbox folder" | Creates a recurring job that stays quiet when there's nothing new || "Process" button | Runs the full pipeline right now |@@ -182,7 +183,7 @@ cd agent-second-brain && bash upgrade.sh```vault/-├── daily/ # Your daily entries (voice, text)+├── daily/ # Your daily entries (voice, text, attachments)├── goals/ # Vision → yearly → monthly → weekly├── business/│ ├── crm/ # Client cardsdiff --git a/README.ru.md b/README.ru.mdindex 1e62240..3576f18 100644--- a/README.ru.md+++ b/README.ru.md@@ -24,7 +24,7 @@## Что делает эта штука-Ты пишешь боту в Telegram. Голосовые или текст - как удобно. Не думаешь про категории, теги или куда что положить.+Ты пишешь боту в Telegram. Голосовые, текст, фото, файлы, пересланные посты - как удобно. Не думаешь про категории, теги или куда что положить.Агент разбирается сам:@@ -93,6 +93,7 @@ Agent Second Brain использует кривую забывания Эбби|----------------|-------------|| Голосовое о звонке с клиентом | Расшифровывает, создаёт CRM-карточку с follow-up || Текст: «идея для Q2 кампании» | Сохраняет в папку идей, связывает с похожими заметками |+| Пересланный пост или фото доски | Сохраняет файл в хранилище, агент сам читает его и раскладывает суть || «Напомни про звонок завтра в 15:00» | Создаёт разовое задание по расписанию, пингует в Telegram || «По будням в 18:30 проверяй папку inbox» | Создаёт регулярное задание, которое молчит, если ничего нового || Кнопка «Process» | Запускает полный цикл обработки прямо сейчас |@@ -182,7 +183,7 @@ cd agent-second-brain && bash upgrade.sh```vault/-├── daily/ # Дневные записи (голосовые, текст)+├── daily/ # Дневные записи (голосовые, текст, вложения)├── goals/ # Видение → год → месяц → неделя├── business/│ ├── crm/ # Карточки клиентовdiff --git a/src/d_brain/bot/handlers/chat.py b/src/d_brain/bot/handlers/chat.pyindex e53b51b..6d59a3a 100644--- a/src/d_brain/bot/handlers/chat.py+++ b/src/d_brain/bot/handlers/chat.py@@ -9,6 +9,7 @@ import asyncioimport htmlimport loggingfrom datetime import datetime+from typing import Anyfrom aiogram import Bot, F, Routerfrom aiogram.enums import ChatType@@ -150,6 +151,74 @@ async def _typing_loop(bot: Bot, chat_id: int) -> None:pass+# --- Media input (photo / document / video / audio / animation / video_note) ---++UNSUPPORTED_REPLY = (+ "Я принимаю голос, текст, фото и файлы. "+ "Этот тип сообщения обработать не могу."+)++_MEDIA_EXTRACTORS = (+ # (kind, attr, default extension)+ ("document", "document", None),+ ("video", "video", "mp4"),+ ("audio", "audio", "mp3"),+ ("animation", "animation", "mp4"),+ ("video_note", "video_note", "mp4"),+)+++def extract_media(message: Any) -> tuple[str, str, str, str | None]:+ """(kind, file_id, extension, original_name) for a media message.++ Photos are a size ladder — take the largest. Documents/audio keep the+ original file name (its extension wins over the default).+ """+ if getattr(message, "photo", None):+ return ("photo", message.photo[-1].file_id, "jpg", None)+ for kind, attr, default_ext in _MEDIA_EXTRACTORS:+ obj = getattr(message, attr, None)+ if obj is None:+ continue+ name = getattr(obj, "file_name", None)+ ext = default_ext or "bin"+ if name and "." in name:+ ext = name.rsplit(".", 1)[-1].lower()+ return (kind, obj.file_id, ext, name)+ raise ValueError("message carries no known media")+++def forward_note(origin: Any) -> str:+ """Human-readable forward attribution, or '' for a non-forward."""+ if origin is None:+ return ""+ user = getattr(origin, "sender_user", None)+ if user is not None:+ return f"[переслано от: {user.full_name}]\n"+ chat = getattr(origin, "chat", None)+ if chat is not None:+ return f"[переслано из: {chat.title}]\n"+ name = getattr(origin, "sender_user_name", None)+ if name:+ return f"[переслано от: {name}]\n"+ return "[переслано]\n"+++def build_media_prompt(+ *, kind: str, rel_path: str, original_name: str | None, caption: str, fwd: str+) -> str:+ """Prompt for the brain: it lives in the vault and can Read the file+ itself (images, PDFs, text) — we only hand it the path and context."""+ name_part = f" (имя файла: {original_name})" if original_name else ""+ caption_part = f"\nПодпись: {caption}" if caption else ""+ return (+ f"{fwd}Пользователь прислал {kind}: {rel_path}{name_part}{caption_part}\n"+ "Прочитай файл (Read поддерживает изображения и PDF; для "+ "видео/аудио опиши по подписи и контексту), сохрани суть в память "+ "по правилам vault и кратко ответь, что сохранил."+ )++# --- Handlers ---@@ -216,17 +285,85 @@ async def handle_chat_text(message: Message, bot: Bot) -> None:settings = get_settings()storage = VaultStorage(settings.vault_path)+ fwd = forward_note(getattr(message, "forward_origin", None))+ text = f"{fwd}{message.text}" if fwd else message.text+# Safety net: save to dailytimestamp = datetime.fromtimestamp(message.date.timestamp())- storage.append_to_daily(message.text, timestamp, "[text]")+ storage.append_to_daily(text, timestamp, "[forward]" if fwd else "[text]")# Log to sessionsession = SessionStore(settings.vault_path)session.append(message.from_user.id,"text",- text=message.text,+ text=text,msg_id=message.message_id,)- await _dispatch_text(bot, message.chat.id, message.from_user.id, message.text)+ await _dispatch_text(bot, message.chat.id, message.from_user.id, text)+++ F.photo | F.document | F.video | F.audio | F.animation | F.video_note+)+async def handle_chat_media(message: Message, bot: Bot) -> None:+ """Handle any file-bearing message: download into the vault's+ attachments and hand the PATH to the brain — it reads the file itself."""+ if not message.from_user:+ return++ settings = get_settings()+ storage = VaultStorage(settings.vault_path)++ try:+ kind, file_id, ext, original_name = extract_media(message)+ file = await bot.get_file(file_id)+ if not file.file_path:+ await message.answer("Не удалось скачать файл.")+ return+ file_bytes = await bot.download_file(file.file_path)+ if not file_bytes:+ await message.answer("Не удалось скачать файл.")+ return++ timestamp = datetime.fromtimestamp(message.date.timestamp())+ rel_path = storage.save_attachment(+ file_bytes.read(), timestamp.date(), timestamp, ext+ )++ caption = message.caption or ""+ fwd = forward_note(getattr(message, "forward_origin", None))++ # Safety net: save to daily with an Obsidian embed+ daily_entry = f"{fwd}![[{rel_path}]]"+ if caption:+ daily_entry += f"\n\n{caption}"+ storage.append_to_daily(daily_entry, timestamp, f"[{kind}]")++ session = SessionStore(settings.vault_path)+ session.append(+ message.from_user.id,+ kind,+ text=caption or rel_path,+ msg_id=message.message_id,+ )++ prompt = build_media_prompt(+ kind=kind,+ rel_path=rel_path,+ original_name=original_name,+ caption=caption,+ fwd=fwd,+ )+ await _process_and_reply(bot, message.chat.id, message.from_user.id, prompt)++ except Exception as e:+ logger.exception("Error processing media in chat")+ await message.answer(f"Error: {e}")+++async def handle_chat_other(message: Message) -> None:+ """Catch-all: never go silent — tell the user what the bot accepts."""+ await message.answer(UNSUPPORTED_REPLY)diff --git a/tests/test_chat_handler.py b/tests/test_chat_handler.pyindex b9b6803..9c3b246 100644--- a/tests/test_chat_handler.py+++ b/tests/test_chat_handler.py@@ -170,3 +170,77 @@ def test_text_when_idle_goes_to_normal_turn(monkeypatch):asyncio.run(chat._dispatch_text(bot, chat_id=10, user_id=1, text="привет"))assert mgr.sent == [(1, "привет")]assert mgr.steered == []+++# ── media input: photo / document / video / audio (v3.0.x) ────────────────+++class _Stub:+ def __init__(self, **kw):+ self.__dict__.update(kw)+++def test_extract_media_photo_takes_largest():+ from d_brain.bot.handlers.chat import extract_media++ msg = _Stub(+ photo=[_Stub(file_id="small"), _Stub(file_id="big")],+ document=None, video=None, audio=None, animation=None, video_note=None,+ )+ kind, file_id, ext, name = extract_media(msg)+ assert (kind, file_id, ext) == ("photo", "big", "jpg")+++def test_extract_media_document_keeps_name_and_ext():+ from d_brain.bot.handlers.chat import extract_media++ doc = _Stub(file_id="d1", file_name="report Q2.pdf")+ msg = _Stub(photo=None, document=doc, video=None, audio=None,+ animation=None, video_note=None)+ kind, file_id, ext, name = extract_media(msg)+ assert (kind, file_id, ext, name) == ("document", "d1", "pdf", "report Q2.pdf")+++def test_extract_media_video_note_is_mp4():+ from d_brain.bot.handlers.chat import extract_media++ msg = _Stub(photo=None, document=None, video=None, audio=None,+ animation=None, video_note=_Stub(file_id="v1"))+ kind, file_id, ext, name = extract_media(msg)+ assert (kind, ext) == ("video_note", "mp4")+++def test_forward_note_variants():+ from d_brain.bot.handlers.chat import forward_note++ user = _Stub(sender_user=_Stub(full_name="Ivan Petrov"))+ assert "Ivan Petrov" in forward_note(user)+ channel = _Stub(sender_user=None, chat=_Stub(title="AI News"))+ assert "AI News" in forward_note(channel)+ hidden = _Stub(sender_user=None, chat=None, sender_user_name="Hidden Guy")+ assert "Hidden Guy" in forward_note(hidden)+ assert forward_note(None) == ""+++def test_build_media_prompt_contract():+ from d_brain.bot.handlers.chat import build_media_prompt++ p = build_media_prompt(+ kind="document",+ rel_path="attachments/2026-06-10/img-120000.pdf",+ original_name="report.pdf",+ caption="квартальный отчёт",+ fwd="[переслано от: Ivan]\n",+ )+ assert "attachments/2026-06-10/img-120000.pdf" in p+ assert "report.pdf" in p+ assert "квартальный отчёт" in p+ assert "Ivan" in p+ # the brain must be told to actually open the file+ assert "Read" in p or "прочитай" in p.lower() or "посмотри" in p.lower()+++def test_unsupported_content_reply_exists():+ from d_brain.bot.handlers.chat import UNSUPPORTED_REPLY++ assert "голос" in UNSUPPORTED_REPLY or "voice" in UNSUPPORTED_REPLY.lower()