Most of the questions a lead business asks in a day are short: how many leads came in, what did they earn, what got stuck. Markmatix is being designed to answer them from your phone, as a Telegram CRM for the people who collect and pass on leads. There will be two tools: an AI assistant, on the web and on Telegram, and a Telegram bot that will be able to do anything the dashboard will do. The AI will prepare and propose. Every action that leaves the platform will be confirmed by a person.
One command layer
The dashboard, the API, Telegram and the AI assistant will all call the same commands, with the same permissions and the same log. No feature will exist in one interface only.
Every command, wherever it comes from, will be written to the action log: who, when, from where (web, API, Telegram, AI), what, and the data before and after. It will be visible in the dashboard and kept for 12 months.
How the Telegram CRM will work
There will be a single platform bot, and it will answer linked accounts only. Linking will start in the dashboard: you will generate a one-time code and send it to the bot. From then on, your Telegram account will have the permissions of your role, and nothing else.
Commands will be the same as in the dashboard and the API. You will be able to give them in plain language, through the assistant, or with buttons. Commands will work in private chats only. In team groups, the bot will post read-only notifications and nothing more.
A bot carrying the publisher's or agency's own name and logo will be available as a white-label plugin. Whether it is included or paid for will be decided in the price list.
Confirmations that match the risk
| Action | What it will take |
|---|---|
| Reading, analysing, preparing | Nothing: these will be free |
| Sends, transfers, erasures, rule changes | A summary with Confirm and Cancel buttons |
| Above a threshold, for example 1,000 recipients or a change to transfer rules | A 2FA code as well |
There will also be a daily sending cap per publisher. In future, each publisher will be able to give the assistant some room, for example sends to fewer than 100 recipients.
Notifications
Each user will pick their own notifications: a daily summary of leads, revenue and chargebacks, alerts, and the results of sends. Some critical alerts will stay on whatever your settings:
- fraud, when a source is quarantined (see lead quality);
- email bounces above the threshold;
- a partner that stops responding to deliveries (see lead distribution);
- the AI spending cap.
What the AI assistant will do
The assistant will be in the core, on the web and on Telegram. It will understand requests in plain language, turn them into commands and answer questions about your data, such as "how much did salary-backed loans earn in October?", the kind of question a publisher of loan leads asks. The figures will come from the ledger described on the revenue tracking page. It will reply in the user's own language.
It will not be an AI sales assistant that chats with your leads: it will work for your team. Pay-as-you-go add-ons will cover the AI quality score, email copy, analysis of the publisher's privacy notice and advanced fraud detection. Each publisher will have a monthly cap tied to their plan. At 80% an alert will go out. At 100% the paid features will stop, while the basic assistant stays on. Usage will be visible in the dashboard.
What data the AI will see
The planned model is Anthropic's Claude, used through the commercial API. It will sit behind a layer that lets us change model or provider feature by feature. Data sent to the model will normally be pseudonymised, meaning no name, email or phone number. Full data will go only to features that need it, and only if the publisher switches them on. Any transfer outside the EU will rely on valid safeguards, such as standard contractual clauses.
Under the AI Act, Redbit will be the provider of the platform's AI systems, publishers using them will be deployers, and Anthropic will supply the model. Anyone talking to the assistant will know it is an AI. Text written by the AI will be reviewed by a person before it is sent. The input, the result and the human confirmation of every AI action will stay in the action log.
If someone steals the bot token
They will not be able to give commands. Every command will have to come from a linked account, and actions above the threshold will ask for a 2FA code. The dashboard itself will require a password plus a mandatory TOTP code, or a passkey, and every sign-in from a new device will trigger an alert by email and on Telegram.
Want to try the bot when early access opens? Join the waitlist.
Updated 3 November 2026