News Agent is a one-agent news desk on your server. It watches your sources: sites, feeds, public Telegram channels, search, product pages. It merges the collected material (repeats of one story become a single record), processes it with AI — writes a summary with citation links, rates materiality, picks out topics and stories, translates — and then raises alerts by your rules and prepares publications that you approve.
What News Agent is
It is not a program where you click buttons and fill in forms. The idea is different: the agent runs the news, and you observe and command. All settings are plain text files in the project folder, written by your AI agent at your request. There is not a single editing form in the interface: every screen only shows the data, and next to every change there is a button that prefills a ready phrase into the chat with the agent.
This is deliberate. Configuring news watching means dozens of small decisions (which sites, how often, by which threshold, where to deliver), and in forms you would be clicking through them for hours. Instead you say it in words: “watch the prices of these three suppliers and tell me when something gets 10% cheaper” — and the agent deals with the details.
Who it fits
- Watching competitors and prices — what changes on suppliers’ sites, when new products and promotions appear. See the Monitors page.
- Collecting industry news — dozens of feeds and sites instead of a manual morning round, with repeats of one story not multiplying.
- Watching brand mentions — search by freshness and domains, separate rules for the important mentions.
- Working for an editorial team or marketing — publication drafts from the collected material, with quote checking and your approval before sending.
- Preparing briefings for management — one scheduled digest instead of a stream of messages in a messenger.
How the work is built
The path of one story looks like this — every step has its own documentation page:
- Sources — you connect sites, feeds, channels and search queries. Fifteen source kinds: Sources.
- Collection — the agent walks the sources politely, saves pages whole and filters repeats: Collection, Politeness & Archive.
- Monitors — assignments for what exactly to watch and at which change to raise the alarm: Monitors.
- Events — every change becomes a card with evidence: what was, what became and by how much: Events & Evidence.
- AI processing — a summary built on quotes, the materiality rating, topics, companies and people, stories, translation: AI Processing.
- Rules — what to do with which events; checked in order, every decision recorded: Rules.
- Alerts — what to send where, with noise protection: Alerts & Noise Protection.
- Channels — email, Telegram, Slack, Discord, Microsoft Teams, webhook, an API call or your own Python code: Delivery Channels.
- Publishing — a draft from the material, your approval and sending: Publishing.
- Feeds — a ready news feed for your site or mailing: Reader Feeds.
- Custom scripts — when the built-in capabilities are not enough, any step extends with your own Python code: Custom Scripts.
All of it is visible on the eight observation screens: Observation Screens. The numbers and search over the collected material are on the Analytics, Search & Exports page.
What you need before the start
- A server or a computer with Docker — Windows, Linux or macOS all work. Details in the Installation section.
- Your Xedant Agent with an AI model — it writes the summaries, materiality ratings and drafts, and also serves as the chat for commands. Without it everything works except AI processing and the chat.
- The management key — the agent uses it to change the settings. It is set once during installation.
- A list of sources — at least a rough one: site and feed addresses, public Telegram channels. The agent will clarify the rest itself.
First sign-in
The first registered user becomes the administrator — there is no separate “default” account. A sign-in holds for 90 days; you will not retype the password. If an external sign-in is configured (single sign-on through a proxy server), users are created automatically on first visit.
First steps
- Connect the agent — set the address and key of your Xedant Agent. After that the chat appears in the interface, and that is where you give the commands.
- Ask for a source — for example: “connect this feed and this site, check once an hour”. The agent creates the settings files and runs the first check.
- Ask for a monitor — “watch the price and the availability of this product, warn me if it gets more than 10% cheaper”.
- Connect a delivery channel — where to send: email, Telegram, Slack. A channel starts delivering only after a test message.
- Look at the events and the inbox — the first results appear on the Events screen and in the “Inbox”.
- Ask for a morning digest — “collect a digest of everything important every day at nine in the morning”. That is how the stream of messages turns into one readable document.
What the product does not do
Honestly and up front, to avoid disappointment:
- There is no manual editor — not a single editing form in the interface. Everything changes in words through the agent, or by editing the files on disk.
- There is no bundled browser — visual page comparison by screenshots is off until an external service is connected.
- There is no X (Twitter), Instagram or TikTok, and from Telegram — only public channels.
- There is no transcription of videos and podcasts and no text recognition in pictures — such material keeps only its descriptions.
- There are no roles or access rights — the product is built for one owner.
Next → Sources