# Your AI, on your account

> Claude Code embedded in your workspace, learning in the open.

The AI that shows its work. Kempt embeds Claude Code, Anthropic's agent, running on your own Claude account, and everything it learns from working with you is a plain markdown note in your home folder that you can read, edit, and undo. It is also entirely optional: Kempt is fully usable without AI.

## Try this

1. Click **AI** in the left rail, or press `Cmd/Ctrl+L`. The panel also lives in the right sidebar as the **AI Chat** tab.
2. Ask something about your notes.
3. When the conversation turns out useful, click the graduation cap button. Kempt harvests what the AI should remember from this session.
4. Open one of the notes it created. Each is a plain markdown file in your home folder, under `_agent`.
5. Open **Learning** in the left rail. The new entry is listed there with everything else the AI has learned.

That round trip is the whole model: the AI learns, the learning is a file, and the file is yours.

## Set up

AI setup lives in **Settings > AI** and is also offered during onboarding. Kempt auto-detects an existing Claude Code install, and a status badge tells you exactly where you stand, for example "Not set up", "Sign-in needed", or "Installed".

1. Click **Install Claude Code**. It runs the official installer ([setup guide](https://code.claude.com/docs/en/setup)).
2. Sign in by running `claude` in a terminal, then `/login`. Click **Open terminal to sign in** and Kempt opens the terminal for you and detects when you finish.

> **Note:** Claude Code requires a paid Claude plan (Pro, Max, Team, or Enterprise). The free Claude.ai plan does not include it.

> **Note:** After a fresh install the badge may read "Installed, restart to finish". Restart Kempt once.

## On your account

The AI runs on your Claude account. There is no API key to manage, Kempt does not resell AI, and your notes are not routed through Kempt's servers.

## Sessions

- Open multiple chat tabs.
- Search your chat history.
- Fork a conversation to explore an alternative without losing the original.
- Rewind to an earlier point in a conversation.
- Sessions survive an app restart.

## Context

Chips in the chat show what Kempt is feeding the conversation: the current workspace, the selected task, the open document. Type `@` to pin extra notes or items yourself.

### Bring in outside folders

This works beyond your notes. Link an outside folder in the file tree (a code repository, a client folder; see [Linked folders](/help/notes#linked-folders)), then mention it with `@`. It becomes an inline chip in your message, the AI receives the path, and it explores the contents with its own tools. Your whole working world becomes context, one mention at a time, and nothing in a linked folder reaches the AI unless you mention it.

## Permissions

Permission modes control what the AI may do: **Ask**, **Edit**, **Plan**, and **Auto**. A separate **Bypass** master switch exists and is off by default. Permission requests, questions, and plans appear as live cards in the chat, so every decision happens where you can see it.

## Safe edits

When the AI updates your plan (a task status, a date, moving items), the change goes through validated app operations rather than raw file rewrites, so your workspace rules are respected. Every AI edit also lands in version history, tagged as AI, with a record of what changed. You can review and revert any of it: see [Your data](/help/data).

## Learning in the open

After a useful conversation, click the graduation cap button to harvest what the AI should remember:

- **Skills**: how you like things done.
- **Memories**: facts about your work.

Each one is saved as a plain markdown note inside your home folder, under `_agent`. There is no hidden state: what the AI knows is a file you can open, correct, or delete. This is a real learned note:

The **Learning** panel in the left rail lists everything the AI has learned. `SOUL.md`, also a plain file, defines the assistant's personality.

Every conversation has a **Plain mode** switch. Turn it on and that conversation ignores everything learned: skills, memories, and personality.

## Scheduled runs

**Scheduled** in the left rail holds recurring AI runs. You create one by asking in plain language in chat, for example "every morning at 9, summarize what I did yesterday". Prompts can include `[[links]]` to notes or tasks, and those are read fresh at each run. Each schedule keeps its own run history.

## Activity

**Activity** in the left rail collects AI sessions and questions that finished while you were away, so long-running work does not need you to watch it.

## Other providers (beta)

Claude is the default. In **Settings > AI** you can switch to Codex (OpenAI), currently in beta; sign in with your ChatGPT account.

If setup gets stuck, see [Troubleshooting](/help/troubleshooting), ask on [Discord](https://discord.gg/ktE8eUeyTj), or write to [hello@kempt.md](mailto:hello@kempt.md).