OmniPub

The essence

OmniPub is a content autopilot and an editorial pipeline for media that need order in the flow of sources. I am building a system where automation takes away the manual routine without killing taste, editorial control and a clear tone of voice.

Status and horizon

Right now this is a prototype-and-product contour that grew out of n8n experiments, a Python aggregator of sources and applied tasks around Telegram channels. The horizon is a content autopilot for Telegram, VK and other channels that saves time, reduces manual routine and helps several media circuits live without the feeling that you spend the whole day ploughing through a feed.

Why now

Content teams and authored media drown not only in production but also in the manual filtering of signals, in switching between platforms and in the constant tuning of tone of voice. OmniPub is needed as an answer to that routine: not a “content generator for the sake of generation” but a system that keeps the editorial frame and helps a project live more sustainably.

The separate problem it grew out of is very everyday and very alive: you physically cannot read the whole feed, and what is useful drowns in the noise. This is even more noticeable when you want to monitor not only Russian-language sources but, say, Japanese Twitter accounts, niche websites, specialised blogs or foreign media on a topic. At that point what is needed is no longer just an RSS reader but a system that collects signals from different parts of the internet, normalises them, translates them, filters them and packages them for the right audience and the specific voice of a media outlet.

Goals

• Automate media research and publication without losing editorial control completely
• Be able to work with several platforms and different tones of voice
• Reduce LLM costs through pre-filtering and the architecture of the pipeline
• Learn to collect rare and international signals on a specific topic, not only to rewrite mass news

What has been done

• A Python aggregator of news sources has been set up
• Filtering by stop words, deduplication and LLM relevance scoring have been thought through
• Earlier applied n8n experiments with publishing to Telegram and Threads have been settled
• The idea of a separate UI and a Telegram control panel has appeared
• There is a pilot sense of how separate parts of the system can live as a Telegram bot or a small control panel

Immediate focus

• Settle the MVP and the list of mandatory modules
• Separate the internal tool from the user-facing service
• Prepare a short one-pager with the economics and the use cases
• Understand where the research system ends and a full-fledged editorial copilot begins

Interfaces and platforms

• a backend that collects and normalises sources;
• an editorial panel for selecting, editing and controlling publications;
• a Telegram control panel for quick decisions;
• a multiplatform publishing layer for Telegram, VK and other channels;
• a tone-of-voice system that makes it possible to maintain several media circuits without rewriting everything by hand.

What matters here is not only the mechanics of publication but also the cultural layer of the product: OmniPub has to be able to tell noise from signal, to take the frame of a particular media outlet into account and not to turn everything into a faceless “neuro-text sludge”. Otherwise there is little sense in automation.

Format and artefacts

• a backend service
• a content pipeline
• a control panel
• automatic publication

Stack and tools

• Python / FastAPI
• PostgreSQL / pgvector
• Celery / Redis
• React
• OpenRouter
• Perplexity API
• n8n

What is needed next

• define clearly the boundary between the internal tool and the future product;
• assemble an MVP that shows the benefit without being overloaded with features;
• set out the economics: where cost falls, where time appears, where quality grows;
• understand which roles are needed for development: backend, UI, product logic, editorial research.

Another strong layer of the project, which I want to keep in the description separately, is connected with studying the language of the channels themselves. There was already a logic of breaking Telegram channels down by categories for that: which words and constructions repeat there, what the average sentence length is, how often the text is broken up by dividers, which patterns actually make a channel “recognisable”. This gives OmniPub not only a technical but also an almost media-linguistic depth.

Skills and roles

• architecture of products and systems
• media automation
• designing editorial pipelines
• optimising the cost of AI circuits

Results / potential

OmniPub gathers my line about automation, content and agentic pipelines into a single product vector. It is no longer just “I know n8n” but an attempt to assemble a system that genuinely helps media projects live and scale. In its strong version it is not about autoposting but about a new editorial infrastructure.

Original (Russian) →