AI agents and automations

Agents for ads, content and documents, invoice reading and voice.

• Personal AI assistant — P.A.I. follows the team chat and records the tasks agreed there, so nobody has to copy them into a separate list. Each task stays visible to the team until someone closes it.
• Meta ads agent — The agent writes Facebook and Instagram ad drafts by market, from the products that sell, and reads the "Not good" notes from earlier ones. Every ad's link is on the market's domain and is set by code. "Create" puts a campaign, ad set, creative and ad into Meta, always paused; "Launch" is a separate button with a confirmation and its own right. The budget belongs to the market and a person sets it; the agent does not choose a budget, pause or delete.
• Ad video and voiceover — With each draft the agent writes the words for the clip's voice (up to 52 syllables, no digits) and an English video prompt. Stress marks come from the system's dictionary; unknown words are returned for fixing, not guessed. The clip is made in an outside tool and added as an address; its post stays unpublished in Meta until you press "Publish".
• AI ad images — With each draft the agent writes an image brief and prompt. Optionally, with a paid key, a button makes an image from the real product photo; a person sees it before "Create". There is no quality check.
• Lead agent — The lead agent assigns tasks to the other agents, reviews their work and sends weak results back to be redone. It tracks AI spend against the budget and stops when the budget runs out. In the operations chat it answers questions and takes instructions.
• Agent permissions and approvals — A check sits between the model's decision and the action itself. The agent can read freely, but any action that sends, spends, deletes or changes a setting stops and waits for explicit confirmation, one action at a time. If no confirmation arrives in time, the agent does nothing.
• SEO agents with approval — After the audit, the agents write a proposal for each task, check each other's work and pass it on for approval. Once you approve, translations into every language start automatically. A single sentence in the chat pauses or resumes the work, and nothing goes live without approval.
• AI agent briefings — When you give a task to an external AI agent, you assemble the briefing here. The module adds the context itself: the company rules and a current map of the API with its available endpoints. The agent starts out with an accurate picture of the system.
• AI product copy — The module writes each description to a fixed template with the same sections for every product, then adapts it per store and language. The text is based on the product's own data. A person reviews and saves it.
• AI invoice reading — You upload a supplier invoice and the module reads its lines, matching them to catalogue items by SKU, by links it has already learned and by name. You confirm uncertain matches by hand, and the module remembers them for next time. Confirming the invoice books the goods into stock.
• Data from photo or PDF — The module reads a photo or PDF (a daily menu, an accounting statement, a photo to identify) and turns it into structured data. It first shows what it read and what it matched it to, without saving anything. Saving is a separate button you press after checking.
• Shipping label reading — The operator attaches a shipping label PDF, and the courier and tracking number fill in automatically. The model reads the page the way a person would, including the number under the barcode. A number entered by hand always takes precedence over the one that was read.
• AI mail triage — With the "AI sorting" checkbox (off until you turn it on) a "Sort" button gives new emails a priority, category, language and short summary, so urgent messages appear first. These are only labels: nothing is done on its own. You translate a foreign-language email with one button.
• Daily AI brief — One screen brings together yesterday's sales, the last week against the one before, the channels, the products, stock and Google search. All numbers are calculated by code. With a button AI writes a short review of what is growing and what is dropping, from these tables only. Writing it by itself every morning is a separate setting, off until you turn it on. Reading it takes a few minutes.
• Ads chat — You ask in plain language about your Google Ads campaigns (what spends without sales, which search terms to exclude). The server builds the table with the numbers of the Overview and Search terms screens, the chat gets only that and has no tool to change anything; the numbers in the answer are checked against the table and any that are not in it are marked. The conversation is kept. The change log has a short summary worked out on opening. Changes to the campaigns are made by a person.
• AI sales analysis — The figures are calculated by code. With the "Write findings" button on a report the model receives the finished table and writes the conclusions in plain language. Every number in the findings is checked against the table, and one that is not in it is shown as an error.
• Scheduled automations — Recurring work runs on a schedule: store syncs, reconciliations, supplier orders, promotions, reports and checks. You don't have to remember to start any of it. The results appear in the relevant module.

Toate modulele