Case · AI integrations and agents · 2026
Wire Room
The conveyor, which turns news feeds into ready-made publications, and the Telegram bot, through which the editor can enter the news manually, bypassing sources altogether. The editorial office that does not sleep: from the moment when the news appeared in the feed or came to the bot, to the publication on the site and in the channel, the system does not ask for a single extra click.
From RSS to the Canal – six stops
Each article follows the same route regardless of the source: automatic collection, rewriting, two quality checks and parallel distribution through channels.
-
Collection
RSS sources are polled and put in a queue for processing.
rss:fetch -
Communications
GigaChat rewrites the text, saving quotes and texture of the source.
auto-rewrite -
doubly
RAG search for Qdrant sifts out replays and events already covered.
NewsRagService -
Monitoring
The final AI-gate checks that it is an independent information drive.
QualityGate -
Publication
The material and cover go to the site through WordPress REST API.
auto-publish -
Newsletter
The announcement with the picture goes to the active Telegram and MAX channels of the site.
publish-telegram
The news can be entered directly from Telegram.
The pipeline does not require a RSS source. The editor opens the bot, dictates the news in his own words - then the bot conducts the text through the same correspondence, the same double check and the same quality gate as the automatic materials, and publishes the result in the selected channels.
Bot: Select sites to publish
Editor: crypto-daily.ru
Bot: Select channels for distribution
Editor: Telegram-channel site
Bot: Send the newsletter.
Editor: Exchange X suspends withdrawal of Y token after attack on bridge
Bot: Copy that. I rewrite, check for takes and prepare a picture.
Bot: The news is out.
Exactly the same path as automatic materials, but without a source at the input:
-
Selection of addressees
The editor notes on which sites and in which channels the news should be released.
selecting_sites → selecting_channels -
Text processing
The text sent goes to GigaChat: title, paragraph structure, picture.
awaiting_text → processing -
Double check
Same RAG test as for RSS materials before queuing.
ready → publication_queued -
Publication and distribution
Posting to the site and mailing to the marked Telegram- and MAX-channels, the bot reports to the chat.
publishing → published
Three layers of noise protection
RAG-dedup
A vector database search finds already published materials about the same event and stops the repetition before going to the site.
Quality Gate
The final AI check before publication screens out advertising and non-independent information drives - regardless of the source of the material.
Rules of the site
Each site decides what categories and channels are allowed to it - the conveyor only publishes where the material is appropriate.
Auto-generation of covers
The image for the article and the Telegram announcement is collected automatically, separately for the tape and for the post card.
Multichannel distribution
One material simultaneously goes to the site, in the Telegram channel and in the MAX - without duplication of manual work.
Manual control
Any stage can be stopped in moderation: the conveyor tells the cause and does not publish the material blindly.
The result
The same conveyor serves the automatic flow from RSS, and manual input from Telegram – the editor does not need to remember two different processes: what got into the system through the tape, what through the bot, passes the same correspondence, double check and quality control.
PHP / Laravel · Redis Queue · GigaChat API · Qdrant · WordPress REST · Telegram Bot API