Post reports and reminders to VK Teams chats as a bot

The agent works in VK Teams through a bot: its first request reads the bot's own data, and after that come the chats the bot has been added to. From there it takes messages, the list of members, admins and blocked users, join requests and information about files. That is enough to post a report or a reminder where the team already sits and to work out what is going on in a work chat.

The visibility boundary here is hard: the bot sees only the chats it has been added to and nothing beyond them. History from before it joined is out of reach as well — events start at the minute the bot appeared in the chat, so a digest for last month in a freshly created chat simply cannot be assembled. VK Teams has no webhooks at all: new messages arrive on an event feed the agent holds open, and it picks the conversation up exactly where it stopped. The cursor there only moves forward — already-read events cannot be rewound to — so the agent never lets one slip past. Until the bot is added, it says plainly that the chat is out of reach instead of returning an empty summary.

On the write side the agent does plenty: it sends, edits and deletes messages, pins a summary at the top of the chat, sends messages with buttons and handles taps through a callback, sends files and voice messages, opens threads on messages, changes the chat title, description, rules and avatar, resolves join requests, removes people from the chat and blocks anyone breaking the rules. Adding members is the one thing not available everywhere: it is a private method that works only in on-premise VK Teams builds and only once a system administrator enables it, and in the cloud the agent says so outright rather than quietly failing. A message in a work chat is seen by everyone, and deletions and removals cannot be rolled back — so before such actions the agent names the chat and the text and waits for confirmation.

The connection is a bot token. It travels as a request parameter, so it has to be kept like a password, and the only way to revoke access is to recreate the bot. Everything the agent does is done as the bot and is signed that way in the chat: people see a robot behind the message, not a colleague, and can reply to it right there. How much moderation is possible depends on the bot's rights in that particular chat — without admin rights it will not delete someone else's message or remove a member, and it reports that shortfall directly instead of trying blindly.