Telegram AI commenting

The module runs comments for matching new posts from selected sources. Before launch, the operator selects accounts, selection rules, and a text prompt.

Launch flow

• Select accounts and sources with available comments.
• Configure post selection and the prompt, then review the launch preview.
• Start the job and review results for every target.

Before you launch

• Warmed accounts: fresh, empty ones are what Telegram restricts first.
• A proxy on every account.
• Access to a model: the shared panel provider works, or bring your own key.
• Channels with comments open: if discussions are off, the channel is skipped.

New-post selection

• All matching — a comment for every new post from the source.
• By keywords — only posts containing the words or phrases you set.
• Random share — a share of new posts gets a comment; the operator sets the share.

Why the module only takes fresh posts

On start the module records what the channel has already published, then waits for new posts, polling the sources every few seconds. It deliberately leaves older posts alone: a comment under a post that just went out sits at the top of the discussion where readers see it, while a reply under yesterday's post sinks down the thread. That gives you a practical rule: if your sources publish rarely, a short watch window can end without a single comment — that is an absence of matching posts, not a failure.

Streams and allocation

One account runs one stream, and your plan limits the number of simultaneous streams. Before launch the panel builds an allocation: how many targets and actions each account is assigned and which sources it watches. You see it before the start, so the load can be rebalanced up front. One account watches up to 10 channels at a time. If there are more channels than the accounts can cover, the extras are not picked up — the answer is more accounts, not a higher pace on the current ones.

Text source

• AI — the connected model writes the text from the selected instruction.
• Templates — the operator's prepared text is used.
• Mixed — the text type is chosen by the configured share and stored in the result.

Writing the prompt

• Describe who is writing and why they read this channel — the reply comes out meaningful instead of neutral.
• Ask for one or two sentences strictly on the post's topic: long comments read as unnatural.
• Explicitly forbid filler phrases like “nice” or “thanks for the post” — those are what give a bot away.
• Do not ask for links: comments carrying links attract reports fastest.

Run progress and results

• The state of each stream: completed actions, errors and skips.
• A per-account and per-target event log with filters and search.
• The text of every sent comment, tied to its source.
• Pause and stop can be applied to a single stream without touching the rest.

Why a target can be skipped

• The source published no new posts during the watch window.
• The post did not match the keywords or fell outside the random share.
• The account used up its daily action limit or is waiting out a pause.
• The model did not answer or returned empty text, so no comment is sent.
• Telegram restricted the action, or the target has comments closed.

Account protection

• The pace profile decides how fast an account works: the calmer the profile, the longer the gaps between comments, and every gap differs a little from the last.
• A daily action limit per account and a night break are configured separately.
• When Telegram asks to wait (FloodWait), the stream pauses instead of pushing the account further.
• Protection can stop a specific action before sending — the results show it as a skip.

Common mistakes

• Launching on accounts that were never warmed up.
• Too many channels per account: the extras simply will not be picked up.
• Identical text from every account — obvious to readers and anti-spam alike.
• Links in comments: better not to leave them at all — that is a direct path to reports and restrictions.