amenbo

A record of your work with AI

See your AI's tasks.

No extra work for you. Amenbo keeps the records automatically. What's to do becomes visible, stays organized — and your AI reads fewer tokens.

Start free See what it does

Works with these AI agents

  • ClaudeClaude Code
  • CodexCodex CLI
  • GitHub CopilotGitHub Copilot
  • CursorCursor
  • Gemini CLIGemini CLI
  • Your own agent

The problem

Make AI agents easier to use — free

Your AI does remember. But that memory lives only inside the AI, where you cannot see it. Amenbo solves that.

You can't see what it remembers

Which decisions it built on, what it forgot — invisible from outside, so you can't trace why.

Long work gets dropped

As the exchange grows, the AI sheds the oldest parts. You stall in the same place again.

Switching means starting over

Move to another AI, or just to another day, and the context doesn't come along. You explain again.

The answer

The record stays on your device.

Not inside the AI — in your hands. So you can read it, it doesn't vanish, and any AI can pick it up.

The Amenbo app icon

Works with

Switch AIs, continue from where you left off.

The substance of the work lives in your device's records. So the AI you use is freely replaceable.

One record at the centre. Every agent reads and writes the same one.

In the app

The AI runs inside it.

Open Amenbo and an AI agent runs in it. Split the panes and each one takes a different job. In the column beside them you read the very code the AI is touching.

Two panes, two different jobs. The column on the right reads the same code.
Two panes, two different jobs. The column on the right reads the same code.
For each pane, pick what it opens with and which model it starts on.
For each pane, pick what it opens with and which model it starts on.
The AI names its own pane, so you can see at a glance what it is working on.
The AI names its own pane, so you can see at a glance what it is working on.

Hands off

No extra chores — the AI writes and clears the tasks.

You write only the first request. Raising tasks and marking them done happens on the AI's side, as it works. Either in a pane of Amenbo, or in your own terminal.

Before

Implement saving
Write the tests
Tidy the meeting notes

You raise them, you clear them. Eventually you can't keep up.

Amenbo

  • Decided the designdone
  • Implemented savingdone
  • Writing the testsin progress

Raised and cleared by the AI. You only asked.

Interfaces

CLI for the AI, GUI for you.

Two windows onto the same record. The AI reads and writes by command; you overlook it on a board.

// for AI agents — CLI

A freshly started AI catches up from the records alone. ↻ replay

// for humans — GUI

Amenbo
You survey it — board, timeline, dependency graph. The AI drives; you watch over it.

You survey it — board, timeline, dependency graph. The AI drives; you watch over it.

Measured

The same answer in about half the steps — fewer tokens.

The records are linked, so the AI follows a few lines instead of rereading the whole history. It also reads 1.4–1.9× less.

Steps to answer the same question (measured, median)

Get started

Just hand it to your AI.

Free to use. Install, pick one folder, hand your AI the request. That's your whole part.

  1. 01InstallmacOS / Windows / Linux.
  2. 02Create a projectName it and choose the folder to manage.
  3. 03Hand your AI the requestCopy it and paste it to whichever AI you use. Any AI works.
Start Free · Windows / macOS

For developers

An AI reads amenbo agent --json and follows intake → list → reserve → execute → finish. Reserving is a compare-and-swap, so parallel sessions never double-work a task.

Projects are isolated; an agent cannot read or write outside its own. Releases are built by public CI with verifiable attestation. Apache-2.0.

$ amenbo agent --json --actor ai

# team

Team

@shiro_doro

Developer

Dev notes →

@SAKANA_ELELEL

Designer

@hidekobayan

Advisor