The GeneralBrain platform

GeneralBrain turns ready-made AI assistants — Claude Code from Anthropic and Codex from OpenAI — into a working team member. You give them tasks in Telegram, they remember what was agreed, carry long jobs to the end and report back. Several paid subscriptions act as one, and every person keeps their own data to themselves.

In testing — opening soon · What this means

What it is and why

Today's AI assistants can do real work: fix a website, prepare a document, go through a pile of files. But out of the box each of them lives in one window on one computer. It forgets last month, stops when the subscription runs out of its usage limit, and serves one person at a time.

GeneralBrain does not build yet another assistant. It takes the best ready-made ones as they are and adds what a business needs around them: a way to talk to them from a phone, a task list with checks, a memory you can search, backup subscriptions, and a clear wall between the data of different people.

One platform, several apps

GeneralBrain is one platform made of several apps. Together they cover the whole path from a message to a checked result. Each app also works on its own: take only the one that solves your problem today and add the others later.

How they work together

You write a task in the Telegram thread of a project. Topics passes it to the assistant and brings the answer back to the same thread.
Router picks the AI model and, if the main subscription has reached its limit, moves on to a backup one. llm-mask hides personal data before anything leaves for the cloud.
When a task is large, Coordinator splits it into steps, runs them one after another, checks each result and reports progress in the same thread.
Knowledge lets the assistant find what was said or sent earlier and quote the source.
Control decides who may do what and keeps passwords and keys out of the assistant's reach. office lets the assistant read and fill Word, Excel and PDF files and build presentations.
Node installs all of this on one or several servers and keeps it running for several people or teams.

How the apps fit together.

The apps

Each app in four questions. All apps share one small common library, which comes with any of them.

Topics

Your AI assistant in Telegram

What it does: Turns a Telegram group into a workplace with an AI assistant. Each thread of the group is one project with its own files and its own conversation history. You write in a thread, the assistant does the work and answers there. Voice messages, photos, documents and reminders work as well. Delivery of messages is a separate building block, so other channels can join. A web panel already works: sign-in through Telegram, screens that show which tasks are running, what the assistant knows and who has access; an administrator manages the platform and adds new members right there. Progress arrives as messages while the work goes on, and one phrase turns them off. For different requests the platform itself decides how deeply the assistant should think: quickly for chat, thoroughly for code. The assistant can also publish a mini-app it built on the internet (for you only, or open) and attach it to the bot.
Who needs it: Anyone who wants to give tasks to an AI assistant from a phone, and small teams that want to see in one group which answer belongs to which project.
What it looks like in practice: On the way to a meeting you send a voice message to the “Website” thread: “the sign-up page shows an error, find out why”. The assistant looks into it, fixes it and replies in the same thread. You read the result when you are free.
Can it be used on its own?: Yes. You need a Telegram bot and a subscription to an AI assistant.

Coordinator

A task list for AI assistants

What it does: Takes a big request and splits it into steps. For each step it starts an assistant in the right project, accepts the result only with proof that it works, and reports as it goes: “step 2 of 5”, a question for you, the final result. Every task is a plain file that a person can read; nothing is hidden.
Who needs it: People who give AI assistants work that takes hours or spans several projects, and do not want to keep the plan in their head or restart things by hand.
What it looks like in practice: In the evening you ask: “update the prices on the website and check every product page”. You go home. In the morning the thread holds a short report: what was done, what was checked, and one question that needs your yes or no.
Can it be used on its own?: Yes. It works with the assistants directly and can report into a chat, a file or your own system.

Knowledge

A memory the assistant can search

What it does: Collects your Telegram conversations, the documents people send to the bot (it reads scanned pages too) and your notes, and lets the assistant search them by meaning, not only by exact words. Every answer points to its source.
Who needs it: Anyone whose agreements live in chats and documents: managers, sales people, project leads.
What it looks like in practice: You ask: “what did we agree with the supplier about delivery in August?”. The assistant finds the message, quotes it and tells you the date.
Can it be used on its own?: Yes. Point it at a folder of documents and any assistant can search it. Each person has a memory of their own that nobody else can read.

Router

Several subscriptions working as one

What it does: Joins your AI subscriptions — Claude, Codex, Kimi, GLM, Gemini, Groq, OpenRouter and models on your own server — into one. When one of them reaches its usage limit, work moves to the next one by itself, and each answer says which model wrote it. Your own programs connect to one shared access point, while the platform decides which model answers and keeps count of the spending.
Who needs it: Teams that pay for several AI subscriptions and do not want work to stop in the middle of the day; developers who want to connect their own bot or program to a subscription.
What it looks like in practice: At three in the afternoon the main subscription reaches its limit. Nobody notices: the next answer comes from the backup subscription and is signed with the name of its model.
Can it be used on its own?: Yes. It is a small service on your server; any program that speaks the common chat format of AI services can use it.

llm-mask

Private data stays private

What it does: Before a text goes to a cloud AI model, it replaces names, email addresses, phone numbers, bank card numbers and passwords with neutral labels, and puts the real values back into the answer.
Who needs it: Companies that handle personal data of customers, and the people responsible for security and data protection.
What it looks like in practice: An assistant answers a customer’s email. The cloud model sees “[[EMAIL_1]]” instead of the address; the customer gets the reply with their real address in it.
Can it be used on its own?: Yes, it is a product of its own and needs nothing else from GeneralBrain.

More about llm-mask →

Control

Rules, keys and schedules

What it does: One place that decides who may do what: for example, only the owner may restart someone’s workspace, and risky actions first ask “Are you sure?”. It keeps the passwords and keys for outside services, so the assistant gets the result of a request but never sees the key. It also starts regular jobs on a schedule.
Who needs it: Owners and administrators of a server where AI assistants work for several people.
What it looks like in practice: A colleague asks the bot to delete a project. The bot politely refuses and writes it down in the log; the owner gets the request with a “Confirm” button, and nothing happens until it is pressed. Every Monday at ten the weekly review starts by itself.
Can it be used on its own?: No, it is the control layer of the platform and works together with the other apps. It pays off most when several people share one server.

office

Office documents for assistants

What it does: One tool that reads Word, Excel, PowerPoint and PDF files — including scans, which it recognises — converts between formats, fills templates without breaking their layout and builds presentations. The original stays untouched: the filled copy appears next to it.
Who needs it: Anyone who works with contracts, offers, price lists and reports and wants an assistant to handle them.
What it looks like in practice: A scanned contract arrives as a PDF. The assistant reads it, finds the payment terms and fills your Word offer template with the customer’s details.
Can it be used on its own?: Yes, as a separate tool for people and assistants on a computer or a server. It brings along the document reader it shares with Knowledge.

Node

The whole platform on one server

What it does: Installs all the apps on one server and runs them for several people or teams. Each of them gets a separate space with its own bot, chats, files, memory and keys, kept apart by the operating system itself rather than by a setting in a program. Node also watches that everything is running and, after every update, checks that nothing got worse. It is built to survive failures: data is copied to a second server continuously, and if the main one fails the second takes over within about five minutes, with never two bots answering at once; this is proven by drills. Any workspace can be restored from the nightly archive with one command, and any update can be rolled back with one command. Watchdogs follow disk and load, and a single view shows all servers.
Who needs it: A company, a team or a family that wants its own AI platform on its own server.
What it looks like in practice: A new employee joins. The administrator writes to the admin bot, answers four questions — name, bot, group, subscription — and a few minutes later the employee’s assistant greets them in their group.
Can it be used on its own?: Node is the assembly of all the apps: it is what you install when you want the whole platform.

Documentation for developers →