> ## Documentation Index
> Fetch the complete documentation index at: https://docs.anonymous.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# 0x0 Nvim Chat

> Neovim chat, inline edits, and agent change review.

`@anonymous-dev/0x0-chat.nvim` is the editor chat client for the server.

It connects to `/ws`, starts agent sessions, streams assistant output, runs inline edits,
and reviews worktree changes before they touch the user's checkout.

## Commands

| Command                   | Purpose                        |
| ------------------------- | ------------------------------ |
| `:ZeroChat`               | Open the 0x0 chat buffer       |
| `:ZeroChatNew`            | Start a new chat session       |
| `:ZeroChatOpen {id}`      | Open an existing session       |
| `:ZeroChatSubmit`         | Submit the current prompt      |
| `:ZeroInlineEdit`         | Edit the selected range        |
| `:ZeroReview`             | Open the change review view    |
| `:ZeroChangesStatus`      | Refresh changed files          |
| `:ZeroAcceptAll`          | Accept all agent changes       |
| `:ZeroDiscardAll`         | Discard all agent changes      |
| `:ZeroAcceptFile {path}`  | Accept one changed file        |
| `:ZeroDiscardFile {path}` | Discard one changed file       |
| `:ZeroCancel`             | Cancel the active run          |
| `:ZeroClose`              | Close the WebSocket connection |

## Setup

Plugin module:

```text theme={null}
zeroxzero
```

Package name:

```text theme={null}
@anonymous-dev/0x0-chat.nvim
```

```lua theme={null}
require("zeroxzero").setup({
  server_url = "http://localhost:4096",
  provider = "codex",
  model = "gpt-5.4",
})
```

## Flow

1. The plugin opens a WebSocket connection to the local server.
2. A chat session creates a detached agent worktree for the current repo.
3. The agent runs in that worktree and streams text back into Neovim.
4. The server checkpoints changed files.
5. The review commands accept or discard those changes.

The checkout stays under user control. Agent edits are staged in server-managed state until
the user accepts them.