It asks first
Anything that leaves your computer waits for a yes. Trust a tool in one chat, or everywhere. You can take it back any time.
Early beta · free
Give Hachi a goal. A small crew plans it, does the work and checks in. It runs on the AI you already have, keeps your memory and keys on your computer, and asks before it sends, buys or posts anything.
I read your Notion board. These three are due first.
You talk to Hachi. Behind it, four helpers split the job, each with one clear role.
Turns your goal into a plan and a budget. Nothing starts until you approve it.
Reads the web and your connected tools, and says where every fact came from.
Drafts the email, the brief or the table. Every change to your files can be undone.
Checks in each day, keeps the goal moving and tells you when something slips.
Anything that leaves your computer waits for a yes. Trust a tool in one chat, or everywhere. You can take it back any time.
Use your own OpenAI, Anthropic or Google key, OpenRouter, or a model on your own computer. Hachi shows the cost before and after each goal.
Memory, chats and keys stay on your machine. No Hachi account. No telemetry.
A morning check-in, a weekly review, or a watch on a page. In the background Hachi only reads; anything else becomes a draft for you.
Notion, Linear, Todoist, Cal.com and your files. Each one is tested against hidden instructions before it joins the list.
Every action goes into a signed audit log. The AI can't edit it or describe it away.
By default, nothing goes to us. This is the whole list.
| Where to | What | When |
|---|---|---|
| The AI you connect | Your messages and what Hachi needs to answer them | When you use it, with your own key |
| Web pages | An ordinary page request | When you ask for research |
| Tools you connect | What their actions need | After you approve them |
| Our update server (EU) | A check for a new version | A few times a day |
Hachi is early. It updates itself, and every bug report makes it better.