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

# Set up Koharu Agent

> Sign in with ChatGPT and configure the integrated project assistant.

Koharu Agent is a Codex client inside the editor. It uses ChatGPT device-code sign-in; you do not enter an OpenAI API key for the Agent.

<Steps>
  <Step title="Open the Agent panel">Open a project and select **Agent** on the right.</Step>

  <Step title="Authorize the device">
    Choose **Sign in with Codex**. In the browser authorization page, enter the displayed device
    code and approve the sign-in for your account.
  </Step>

  <Step title="Choose a model">
    Return to Koharu and wait for the account and model list. Select a Codex model and reasoning
    level in the Agent properties. More reasoning can help with ambiguous edits but takes longer.
  </Step>
</Steps>

Only one sign-in attempt can run at a time. Cancel it before starting another.

## What context is shared

The Agent starts with the open project's semantic state. It renders and attaches a page image only when it invokes visual page inspection; it does not automatically attach every page.

<Note>
  Agent sign-in and translation-provider credentials are separate. ChatGPT sign-in does not
  configure the pipeline's OpenAI provider, and an OpenAI API key does not sign in the Agent.
</Note>

## Sign out

**Sign out** cancels active work, clears the current Agent session, and removes the Codex account authorization used by Koharu. Agent configuration is saved separately from translator settings.

<Card title="Work with a project" icon="comment" href="/en/agent/projects">
  Give the Agent a precise scope, review its edits, and use project history.
</Card>