Content moves like an assembly line. Your editors control every step.
Genix is a cloud platform for media. Assemble a route from ready AI steps — rewrite, fact-check, headlines, cover, voice-over, publish — without a development release. Every run is visible: what came in, what went out, where it waits for a person.
A media unit is a workshop with its own production lines.
A media unit in Genix is not just "a site" or "a station". It is a self-contained production workspace with its own routes, rules and launch points. Think of a workshop: its own profile, its own specialisation, several production lines inside. The sharper the specialisation, the more predictable the output.
A pipeline is the route through the workshop.
Which stations, in what order, where the system acts on its own, where it waits for a person. Built once, used by the whole desk — manually, from a dashboard, on a schedule, or triggered by live air.
A content item is a cart moving along the line.
Not just the finished article. A cart that travels the line and collects artefacts: text, voiceover, cover image, working data, intermediate versions. The final product is the result of a route, not a file someone made somewhere.
A run is one journey of one cart along one route.
Not abstract "AI generation" — a specific process for a specific item, with a specific starting point. The same route gives different results depending on what enters it, so the launch point matters.
A step is a station on the line.
Takes an input, performs one action, hands the result on. Steps are isolated, which is why a local fix stays local.
The coffee-machine principle.
Beans in, settings chosen, coffee out. If the coffee is wrong, you change the setting on the stage that is wrong — not the whole machine, and not by pleading with the previous stage. Genix works the same way: each step owns its part of the result, so corrections are surgical.
Six families of station
Data
News feeds, semantic search for related material, full text from any URL, video transcripts from a link, any external REST API, AI relevance scoring, tag selection, splitting a transcript into stories.
AI
Rewrite, summary, script, social announcement, translation with adaptation, headline, fact-check, structured fields against a schema you define.
Media
Text-to-speech and audio mixing, image generation, templated covers and banners, stock image search, avatar video, short AI video, document generation, audio transcription.
Flow
Nested sub-pipelines with typed input/output contracts, informational blocks for the operator, branching.
Input and human decision
Manual field edits, waiting for your text or files, an explicit approval before continuing.
Delivery
Site CMS, messenger channels, social platforms, audio digests, radio station cards, cloud drives, object storage.
56 step types in the catalogue. Which of them exist in your contour depends on which integrations are connected. That is configuration, not a bundled "everything included".
Editors build the route, not engineering
Routes are assembled on a canvas as a visual flow of steps: order, conditions, nested sub-pipelines with typed contracts.
Prompts live in libraries, per step type — globally and per media unit.
An AI profile (provider, model, parameters, limits) is a setting, not something hard-coded into the product.
A route can be cloned and re-pointed at another media unit. That is how a new brand gets a production process without a development project.
When an editorial task needs a new behaviour, it usually needs a prompt and a setting — not a release.
Where the route stops, and who decides that
Approval gate.
Nothing continues until a person confirms.
Manual field edit.
The route opens specific fields for editing and resumes with the edited values.
Input stop.
The route waits for your text, files or links, and can transcribe attached audio before continuing.
Automation level is per route.
Raise it where the scenario is proven, lower it where risk is higher. The newsroom sets this, not the vendor.
Every published piece has a history
Source → steps → prompts → editor's approval, recorded per item. Step statuses, inputs and outputs of each step, retry history, and which run produced which artefact.
Traceability is not a feature bolted on later. It is a consequence of describing the whole process as a route of discrete steps from the very beginning. A tool whose unit of work is "one generated text" cannot retrospectively show a chain that does not exist in its data model.
We record what the system did, from which source, with which prompts, who started the run and who published the result. There is no per-step audit trail of exactly who approved which gate and when. Do not build a legal argument on this line.
The system hears the story before a person does
Live audio is transcribed continuously.
A classifier detects where one story ends and the next begins.
Speakers are separated.
The route starts on its own, with the segment and its context as input.
Transcription itself is common — 78% of newsrooms already use AI for transcription and translation (WAN-IFRA · FT Strategies, 2026). The part we have not found in open practice is the full automatic path from live speech to a published piece. That is our own market observation, not an independently verified claim.
A new title is configuration, not a project
One platform runs production for many media units simultaneously.
Each unit has its own routes, prompts, provider keys and spending limits.
A new vertical or a new language version is a cloned and re-pointed configuration.
Isolation today is at media-unit level inside one deployment.
Verification that can stop the line
The AI output is compared against the source material.
Claims that are not supported by the source are flagged.
Depending on the route's setting, publishing either waits for the editor's confirmation or continues with the remarks recorded. The newsroom chooses per route.
The documented failures of 2025–26 share one cause: a skipped editorial check. A verification tool that lives beside the process is easy to forget. A step inside the route runs as part of production — and the run page records which verification setting was active for each item.
One item, several outputs
Article
Publishes to your CMS through its API, with an HTML body matching that title's editorial standard and tag selection. Connectors exist for several CMSs today; yours is connected during rollout
Audio / podcast
Text-to-speech, segment assembly, stingers and loudness normalisation into a finished track
Radio digest
An audio digest into a delivery service, or a station card filled in
Video
Avatar video, short AI video, or a render from a video template
Social and messenger
Text, photo, audio, media album, document; feed posts, carousels, stories, reels, threads
Documents
Word file or a document on a cloud drive
Which outputs exist in your contour depends on which integrations are connected. This is configuration, not a bundle.
Genix runs beside your stack, not instead of it
Sources:
news feeds, semantic search, full text from arbitrary URLs, video transcripts, and any external REST API through a generic HTTP step.
Models:
OpenAI, Anthropic, Google / Vertex AI. The model is a parameter of an AI profile.
Voice and audio:
speech synthesis, speech recognition, audio mixing.
Images and video:
several image generators, templated covers, stock search, short video and avatar video.
Delivery:
site CMS, messenger channels, social platforms, audio digests, cloud drives, object storage.
Automation:
scheduled runs, runs triggered by live air, and integrations through an automation layer.
The well-known AI vendors are not our competitors. They are stations inside the line. Put the model you consider right into the route; replace it when a better one appears.
You see the cost per unit before you see the invoice
Hourly accounting of every paid AI call: tokens, speech characters, images, video minutes, call count.
Limits on those five counters, set globally and per media unit, enforced by the system.
Other paid external calls — speech recognition, web search, page retrieval — are metered and governed by the contract rather than by a hard cap.
Cost visible per brand, so a single scenario does not quietly absorb the whole budget.
Every paid AI call is metered hourly, and the cost dashboard shows spend per media unit.
The boundaries, stated plainly
Genix ends at publication to your CMS or channel. Reader personalisation, reader-facing chatbots, community and comment moderation, monetisation and paywalls, content litigation, deepfake detection of third-party media, and training our own models — all outside the product.
We build the contour that any model plugs into. We do not train the model.
See it on your own material
The fastest way to evaluate this is to take one of your real stories and run it through a route. We will show you where the system works alone and where it stops and waits for you.
Book a call