KödChat¶
KödChat puts Claude Code and Codex conversations beside your project, files, editor, and terminals. It runs the official CLIs on your Mac; ködade does not proxy model traffic or hold provider credentials.
Start a chat¶
Install and authenticate Claude Code or Codex first. Then:
- Add or select a project.
- Expand the project under KödChat in the sidebar.
- Select + beside the project.
- Choose the provider, model, and access level in the composer.
- Enter a message and press Enter. Use Shift-Enter for a new line.
The first message becomes the thread name. Each later turn resumes the same CLI conversation when the provider supports it.
Choose an access level¶
Access is selected per thread and maps to the provider CLI's own controls:
| Level | Intended use |
|---|---|
| Plan only | Read the project and propose changes without writing. |
| Standard | Read files, edit the project, and run commands. |
| Full access | Let the agent act without permission prompts. |
Review the selected level before sending. KödChat does not add a sandbox beyond the controls supplied by the CLI.
Add context or open a terminal¶
Drop files onto the chat pane to add their paths to the next message. The agent still decides whether and how to read them under its selected access level.
Use the terminal control in the KödChat header when you need an interactive shell, a provider login flow, or a CLI that does not yet have a chat adapter. Claude Code and Codex are the supported KödChat providers in the current public release; other detected CLIs remain available in a terminal.
Understand transcript storage¶
KödChat saves each thread as a local JSON document in ködade's application-data directory. Closing a chat thread deletes its transcript document. Transcript text is not copied into KödMem activity records.
See Local data and privacy for the storage boundary and Agent CLIs for installation and authentication.