AiHummerAiHummer AiHummer
Start free Sign inMy account

AI ASSISTANT · FOR DEVELOPERS

The code is clear. The reasoning isn't.
The project remembers why decisions were made.

Telegram, WhatsApp, VK, MAX. You hand it your decisions, notes and specs — then ask why something was built the way it was. Coming back to a project after two weeks stops costing you half a day.

This is the project’s memory, not a second developer. It doesn’t write code for you, and it enters the repository only where you granted access, and only to read. It answers from your decisions and says where it took the answer from.

1 590 ₽Free

10 days. After that — 1 300 ₽ per month when billed annually.

Get startedSee the live demo
Moscow
Innovation
Cluster
My projectonline
Why did we move from a queue to webhooks?
Checking your decisions…
Decision from March 12: exactly-once delivery cost more
REMEMBERS FOR THE PROJECTdecisions · 9agreements
Build a list of the decisions made on this module
Done — the file is above: 9 decisions with dates
DECISIONS MADETable, 9 rowsSent as a file into this same chat

Fits into what you already use every day

Full integrations catalog →Connect your own →

Services in this ribbon: 38. Which of them are open depends on your plan. View all plans →

  • Telegram
  • Pocket Agent
  • SIP
  • WhatsApp
  • Google Calendar
  • GitLab
  • Asana
  • Gmail
  • Google Tasks
  • Microsoft Teams
  • Obsidian
  • Outlook Mail
  • Telfin
  • Telegram
  • Pocket Agent
  • SIP
  • WhatsApp
  • Google Calendar
  • GitLab
  • Asana
  • Gmail
  • Google Tasks
  • Microsoft Teams
  • Obsidian
  • Outlook Mail
  • Telfin
  • Einstein
  • MAX
  • Avito
  • Email
  • Website widget
  • Битрикс24 CRM
  • ClickUp
  • Google Contacts / People
  • Jira Cloud
  • Microsoft To Do
  • OneDrive
  • Signal
  • Todoist
  • Einstein
  • MAX
  • Avito
  • Email
  • Website widget
  • Битрикс24 CRM
  • ClickUp
  • Google Contacts / People
  • Jira Cloud
  • Microsoft To Do
  • OneDrive
  • Signal
  • Todoist
  • Mobile app
  • Bitrix24
  • VK
  • Telephony
  • GitHub
  • Яндекс Диск
  • Discord
  • Google Drive
  • Matrix
  • Notion
  • Outlook Calendar
  • Slack
  • Mobile app
  • Bitrix24
  • VK
  • Telephony
  • GitHub
  • Яндекс Диск
  • Discord
  • Google Drive
  • Matrix
  • Notion
  • Outlook Calendar
  • Slack

What it costs to get back into your own project

09:40
incoming issue overnight
11:15
what to fix first
14:30
there was a decision, but where
17:50
rewrote it, then remembered why

The code is readable, but the intent of — is not. The most expensive thing is not the forgotten function, but the forgotten reason: why it was done this way, what has already been tried, and which of the nightly tasks is really on fire. All this is recorded — in the issue, in the panel, in the note, — is simply not in one place.

Three works, and none of them are about correspondence

A developer's morning is eaten up not by clients but by triage: twelve incoming issues, two boards with your tasks, and the question “why is it done this way”. Three different jobs, and under each illustration it says which module handles it.

issue12
Overnight inbox
CSV import crashesrepair
Same as #204double
No repeat stepsask
Feature requestnot a bug
sorted through GitHub and GitLab — you can see where to start the day
Today

Twelve issues per night, one of them is on fire

The morning goes on reading everything end to end and separating duplicates from a real bug report by hand.

With the agent

Triages the inbox before you arrive

He reads the inbox, notes duplicates and those where there are not enough repetition steps, and tells where to start the day.

What is on you
AIH-412 is waiting for your reviewJira, third day
Reply about the API contractTodoist
Update the migration schemaoverdue
Jira · Todoist — one list, no switching between boards
Today

What hangs on you — in two panels at once

Some tasks are in the team tracker, some in your personal list. To see the whole load you have to open both and add them up in your head.

With the agent

Builds your list from both boards

One list: what is awaiting your review, what is overdue, and what you created for yourself and forgot to open.

ADR004
Why it was done this way
ADR-004: queue rejected12.03
We agreed on idempotencyin May
“Maybe bring the queue back?”today
taken from your ADRs in Notion and your notes in Obsidian
Today

The code is readable, but the intent of — is not

Why this path was chosen and what has already been tried is known only to whoever was in that discussion. Everyone else starts it over.

With the agent

Answers from your own ADRs

To “why is it done this way,” he responds with your own ADR and note, names the date and says if you have already gone this route.

One morning you don’t lose costs more than a month of the assistant

Try it on your own project →
6 hours
Lost every week

In the example: trying to recall the reason for a decision, hunting for it in a thread from three months ago, rewriting it, and halfway through remembering why it was done differently.

1 300 ₽
Assistant for a month

When billed annually. Answers in your own messenger, and from your phone too.

What gets counted here is hours, not revenue

You already have the project context — it’s just stored in your head. No use to you? Turn it off in one click.

What you no longer have to remember

From your materials

Notes, specs, write-ups, what you agreed with the client.

Answers right away

In the same messenger you already write in.

Keeps the reasons behind decisions

Not just what was done, but why that option won.

It doesn’t touch code

It has no access to the repository — you never granted any.

Live in an evening, no developer needed

Start today →
5 minutes

You connect a channel

Telegram, WhatsApp or the website widget — by the instructions in the dashboard, no developer.

Telegram WhatsApp Widget
one evening

You upload your documents

What is not in your materials it says so, rather than composing it.

PDF Price List 2026DOC Policy… Warranty
in the morning

Answers when you ask

You see every conversation in the dashboard and in the app. Don’t like an answer? Fix it as plain text.

4 chats overnight

What it’s worth to stop rebuilding context

1 590 ₽/mo

or RUB 15,900 per year

  • 1 AI agent
  • Telegram, WhatsApp, VK, MAX
  • Mobile app
  • Your notes, specs and decisions
  • Lists and tables as a file straight into the chat
  • Your discussions, remembered
Get started

The trial period is 10 days.

Need your own server?

How the install works →

Your own server or a virtual machine and PostgreSQL — no Docker needed. It installs with a single command, everything else is set up in the web interface.

  • Installs on Linux — a tarball and systemd, updated with one command
  • Conversations and documents stay with you and never leave
  • Runs on a local model — it can work with no internet at all
See how it installs
the Community plan
0 ₽

Forever, no card. Telegram, web chat and the website widget, one agent, your own server. The rest of the channels come with the paid plans.

Get it free

Questions

Will it write code?

It doesn’t write code for you and doesn’t make the decisions — that is your work. It doesn’t change files: the product has no file-writing tool at all, on any plan. It enters the repository only where you granted access, and only to read; commands are off on the cloud plan, and on your own server you switch them on yourself.

How is this different from searching the repo?

Search finds a line, not a reason. The assistant answers the question “why was it built this way” — from your notes and discussions, which is where that reason actually lives — and names where it got the answer. It reads code as text, but the value here is intent, not finding a function name.

What do I have to give it to make it useful?

What you already have: module specs, notes on decisions made, the thread with the client, diagrams. After that you add a line right in the chat whenever you decide something. What you never gave it, it doesn’t invent: it simply says it doesn’t have that.

Where do we write to it?

Right where you already are: Telegram, WhatsApp, VK and MAX. Every thread lands in one inbox, and there’s a mobile app too — you can ask about your own decision from the subway.

What is the file it sends?

A table — CSV or XLSX — that it builds itself and sends as an attachment right into the chat. That’s how lists of decisions made, module inventories and lists of what you have already tried go out. Handy when you hand a project to someone else.

Does it remember what we already discussed?

Yes. It pulls up your earlier messages and answers with them in mind: which road you have already been down and how it ended. The limit is an honest one — if you decided something out loud on a call and it never reached the chat, it doesn’t know.

Will my material be used to train a model?

No. Your files and conversations are used only to answer you. You connect the model yourself — it can be a local one, on your own hardware, and then nothing leaves at all.

Do I need a separate assistant for each project?

This plan includes one assistant. For a single active project that is enough, and it tells neighboring material apart by the names you gave it. If you would rather split projects across separate assistants, there is room for that on the next cloud tier — the pricing page says exactly how much.

What has to be done before the first answer?

Connecting a model is the first step in the dashboard, and you do it once. Until a model is connected the assistant stays silent: the thinking happens in the model, not on our server. After that you connect a messenger and hand over your material.

Can I see it before paying?

Yes. The demo sandbox is open without sign-up — login and password below, it resets every hour.

Don't take our word for it — check for yourself

Open the sandboxStart in your account

This page is about one job. Everything about the whole product — how it works, pricing, channels and security — on the home page.

The same assistant — and not only in code

Same mechanics, different reasons to write: a sysadmin sheds the same questions from colleagues, a home server explains its own setup, a researcher works through a mountain of sources.