AI assistant
The AI assistant, with the right context
Full or side-by-side chat, attachments, model, effort, approvals and change review.
In this guide
Connect Codex in settingsPrepare the requestModel and permissions are separate choicesKeep the thread of your workConnect Codex in settings#
Install Codex on the same PC and sign in to your account. If it is already configured, check the connection in Code Weaver settings. AI requests are processed online by OpenAI.
- In settings, find Codex and ChatGPT account.
- If you have not signed in yet, choose Sign in with ChatGPT and complete sign-in in your browser.
- Return to Code Weaver and select Check connection.
Prepare the request#
Place beside editor moves the same conversation to the right of your sources. Full view returns it to the central workspace. Draft, history and the active request are shared. You can pin files and instructions, attach images and organize chats with tags, folders and branches.
- Open the assistant and choose a project and conversation.
- Use @ or Add context to select files, folders, selections or symbols.
- Review included and excluded items; open files can use their current editor buffer.
- Choose the mode, model, effort and approval profile, then send.
Model and permissions are separate choices#
The Model selector shows the options available to your Codex account. Effort adjusts reasoning where supported. Ask for approval, Approve for me and Full access instead govern native operations. Approve for me can reject requests; it is not unlimited consent.
Chat provides answers; Plan prepares analysis without applying changes; Agent can use tools according to the chosen profile. Review files and hunks before applying a proposal. Checkpoints support controlled recovery of covered files but do not undo commands, network requests or database effects.
Keep the thread of your work#
Conversation tools includes answer comparisons, activity and recovery, archives and backups. Editing a question or regenerating creates a version while retaining the original. Pausing the queue stops subsequent starts but does not interrupt the active turn. Refresh from Codex and compaction are explicit actions.
Ordinary ChatGPT cloud conversations are not automatically synchronized with local conversations. A conversation’s token count is not the remaining quota of your ChatGPT subscription.