Checked against the implementation on 24 October 2026.
This guide explains Deskgram 2 AI commenting: channel selection, generated comments, delays and delivery checks. See the module overview for its interface and capabilities. Start with your own channel with an open discussion and an account allowed to post there.

What the module does
The task watches selected channels for new posts and sends comments to their discussions. Enable AI for text based on a publication; use the message editor for a prepared comment.
Configure text and AI
Open AI, enable generation and choose provider, model and comment style. For a custom style, enter instructions such as “Give a short comment relevant to the publication, without invented facts or unsupported promises”.

See the AI component guide, custom-provider setup, and service guides for Mistral, OpenRouter and io.net.
Save the key in global settings. Enabling AI and loading models do not prove generation works. If a model is unavailable or AI handling alternatives are exhausted, the module reports an error and may stop commenting rather than necessarily switching to a prepared template.
Check availability, balance and cost with the provider. Without AI, fill the editor and independently test sending the prepared message.
Choose channels and delays
In General, select all account subscriptions or an explicit channel list. Start with one named channel. If needed, join the channel first.
| Setting | Meaning |
|---|---|
| Comments per account | Work volume for each account |
| Delay before sending | Range in seconds before a comment |
| Break between comments | Interval in seconds between comments |
| Exclusions | Sources to skip |
| Delay between account starts | Offset between starting accounts |
The lower bound must not exceed the upper one. Comment count is task volume, not Telegram's daily limit. For FloodWait, inspect the requested waiting period in the log.
Configure editing only if needed

The Messages tab sets replacement content and the delay before changing a comment already sent. Leave editing off when not needed. Test a comment, wait for the configured interval and compare before/after text.
Add the embedded autoresponder if needed
For incoming messages after publication, enable the embedded autoresponder and configure text or AI, duration and reply count. Its settings are covered in the component guide.
Keep it off while testing comment delivery. Check discussion replies and private incoming messages separately.
Review additional settings
- Automatically join discussion chats: attempts joining when posting requires membership. Private discussions and send permissions still need separate checks.
- Reply to replies: handles replies to a comment. Test a fresh reply from another account.
- Message SpamBot on flood: checks account status through @SpamBot; it does not shorten FloodWait.
- Automatically unsubscribe: leaves sources under the selected condition, including posts without commenting available. Inspect which channels may be left before starting.
Enable extra functions individually so you can identify which setting caused a changed result.
Start with a new test post
- Select one connected test account.
- Inspect channel, discussion access, text, delays and task volume.
- Start the task and publish a new post in the test channel.
- Wait for the configured delay, inspect the discussion and locate the send in the log.
Without new posts, the task may simply be waiting. An existing publication does not replace a fresh-post test.
Check statistics

Compare successful sends with actual discussion comments. If none appears, check:
- A new post appeared while the task ran and its channel belongs to the sources.
- Discussion exists and the account may write.
- No delay is active and configured volume remains.
- The prepared message exists or AI credentials/model work.
- The log's actual error: Telegram, network, session or AI service.
Common questions
Can it work without AI?
Yes, with prepared editor text. To generate a comment from post content, AI is needed.
Why fewer comments than new posts?
Some sources have closed discussions. Also inspect exclusions, delays and delivery failures. Processed-post count and published-comment count are separate metrics.
Are editing and autoresponder required initially?
No. Verify one comment first, then add extra functions separately.
Related material
- AI commenting overview.
- Download Deskgram 2 and plans.
- Detailed commenting article.
- Program interface.
- Module on GitHub.
- Other GitHub modules: private messaging, autoresponder, audience parser.
The retained screenshots show Russian controls from an earlier interface. Current labels may differ.