# Personal AI Assistant A context-aware assistant in Telegram, Slack or Discord that knows your business and runs on your own server. Your own AI assistant in Telegram, Slack or Discord, on a server you own. It knows your business, connects to Google Workspace, Atlassian, Git and anything with an API, and you decide who on your team gets what. Fixed price, installed by me. **Status:** beta **Category:** assistant **Price:** $590 **Tags:** AI Assistant, Self-hosted, Integrations --- Your own AI assistant, in the chat app your team already opens every morning: Telegram, Slack or Discord. It knows your business, it reaches the tools you connect, and you decide who on your team gets how much of it. It runs on a server you own, under the name and the face you give it. ## A colleague you write to It answers from what your company actually knows, not from a blank page. Tell it "always do it this way" and it writes the rule down. Walk it through a multi-step job once and it saves the steps under a name you can call later. Every chat it sits in keeps its own history, searchable by keyword and by date. Rules and workflows live as plain files in your workspace, so they outlive any reinstall. In your own chat it does the work, not just the answering. It runs commands, edits files, searches the live web, and works in the services you connected. Photograph an invoice and ask what it says. Forward a contract and ask what you signed up to. Say it out loud instead of typing, and it shows you the words it heard before it acts. Ask for a plan first and it writes one and waits. Approve it, or tell it what to change. Nothing from that plan runs until you tap. ## Connected to the tools you already use A connections layer between the assistant and the services you run, powered by [integrations.sh](https://integrations.sh/): Google Workspace, Atlassian, GitHub, GitLab, Notion, HubSpot, Stripe, and almost anything else with an API. Connect a service once and every question, reminder and scheduled job can use it. It reaches nothing you have not connected yourself. Only the access levels you pick get that reach, and a group chat gets none. Show it the steps once and that job replays without the model, for pennies. ## You decide who gets what An access layer sits between your team and the assistant. You are the owner, set once when it is installed, and that seat cannot be granted or talked into changing from a chat. Below it you hand out two levels. A trusted few get what you have, minus the ability to add or remove people. Everyone else gets a helper that answers questions and drafts work in a private space of its own, but cannot reach a single one of your services. A group chat always gets the weakest level, whoever is in it. Hire someone, add them by replying to their message, or just tell the assistant to give them access, and on day one they have a helper of their own. One tap changes their level or takes it away. Every change is written to a log with who made it and when. ## In your team's group chats Say its name and it answers. Ignore it and it stays quiet. You pick the name, and anything it cannot confirm was meant for it gets no answer at all. It keeps what it reads. Three weeks later somebody asks what the supplier promised, and the answer is there. Each group builds its own history, plus a short sheet of facts it maintains for that group. It never writes down a guess or a password, and one group can never read another's. In a group it can only answer. It cannot touch a file or open one of your services from there, and that holds for you as much as for anyone. Add it to a chat you are not in and it leaves before saying a word. Everyone but you gets 10 answered messages a day and 100 across the group. ## You never have to type A question, a reminder, a whole scheduled job, all dictated. Reminders take plain words: "Remind me about the invoice on Friday" is all of it. The list puts Done and Postpone on every item, and snoozes by 15 minutes, an hour, four hours, a day, or a time you dictate. They arrive when they are due, once, and never twice. ## Work that runs while you sleep Set up a job by describing it once. Monday 8am, last week's sales in your chat. Every night, a check that the backup ran and a message only if it did not. Friday 5pm, every invoice still unpaid. Any schedule you like, in your own timezone. The task list shows each job with its next run and how the last one ended, with pause, delete and run-now on each. A failed run tells you once, in one sentence, then tries again on schedule. Only you can set up a job, because a job acts with your authority. ## On your own server It sits on your server and calls out to your chat app. There is no door into it from the internet, and no shared database with other companies' data next to yours. It restarts itself after a crash and comes back after a reboot, and an update does not kill work in progress. Everything it knows lives in one folder, so copy that folder and you have the lot. Nightly database backups keep 14 days of snapshots. Every run appends one line to a log with who asked, which model answered, how long it took, and what it cost. Voice notes leave once, to be turned into text. Your message text goes to the model that writes the answers. Beyond that, only what you point it at: a web search you asked for, or a service you connected yourself. ## The install $590, once, and I do it. Your server, your bot in Telegram, Slack or Discord, your name and avatar on it, your services connected, your team at their levels, and your first scheduled jobs built with you in the room. It works with the ChatGPT plan you already pay for, connected once from your phone with a code posted in the chat. Past the $590 you are paying for that plan and a small server of your own. Connecting outside services runs through integrations.sh, which bills separately, and everything else works without it. ## Links - Product page: https://andrey-markin.com/products/personal-assistant - All products: https://andrey-markin.com/products.md - Buy this offer: https://andrey-markin.com/offers#business-ai-assistant - Contact: https://andrey-markin.com/#contact