Fixed

• Replit chats and team workspaces now work. Nexpath only recognised project pages at
replit.com/@user/project, so a prompt typed in a chat (replit.com/chats/…) or inside a team
workspace (replit.com/t/team/…) was ignored and no suggestion appeared — including the first
prompt typed on the Replit home page, which was held for a project that never arrived. Those
pages are recognised now, and each project and each chat keeps its own session. Nothing
changes on Lovable or Bolt.

• With a Nexpath token, the improved prompt is now composed by the model, the same way it is
with your own OpenAI key. Until now the check that decides whether a model call is allowed
recognised only OpenAI keys, so token users were quietly given the simpler template-built
suggestion instead. Nothing changes in what the extension sends or where; the call that was
always meant to happen now happens — so a suggestion uses more of your account credit than it
did in 0.1.54.

Changed

• The Nexpath token card in Settings now links straight to your account page, where the token is
shown.

• The extension and the Nexpath CLI now read one shared definition of a valid token, so they can
no longer disagree about a token the service issued.

Housekeeping release. Nothing changes in what the extension does or what it sends.

• The privacy policy has been rewritten into a clearer, conventionally structured
document covering both credential modes, the rating prompt, and what the Nexpath
service records for your account. It is published at the same address.
• An internal consistency fix between the extension and the Nexpath CLI. The set of
content-free action names the extension records is unchanged.

Nexpath token support — you can now run Nexpath without an OpenAI API key.

Added
• Nexpath token. Create a free Nexpath account and paste your token instead of an OpenAI API key; Nexpath then routes its calls through its own service using your account credit. The bring-your-own-OpenAI-key path is unchanged and still works exactly as before.

Changed
• The settings page has been rebuilt around the simpler setup: the Nexpath token card comes first, with step-by-step instructions and a link to your account page.
• A held prompt can no longer be lost. If the page itself abandons the request while a suggestion is open, your original prompt is restored into the composer instead of disappearing.
• A resend that differs only in spacing or line breaks is no longer treated as a new prompt.
• The data-collection notice shown when you add Nexpath to Firefox now lists only what the extension actually requires.

Faster delivery, and no clipboard permission prompt on Chrome.

• The suggestion arrives sooner — the send check now samples densely for the first second instead of waiting out a fixed interval. The overall time limit is unchanged.
• Chrome no longer asks to see your clipboard: delivery on Bolt uses a direct text insertion rather than a synthetic paste event.
• Replit delivery goes through the editor itself, so the prompt arrives in one step regardless of length.
• Fixed a case where a prompt could be sent twice, and one where a multi-line prompt was reported as undelivered when it had arrived.

Permissions, supported sites and privacy behaviour are unchanged. An OpenAI API key is still required.

The suggestion now arrives before your prompt is sent, not after the agent replies.

• Submitting a prompt briefly holds it while Nexpath prepares a stronger version, and the panel offers both. You choose which one is sent.
• Choosing "Use original prompt" sends it immediately; the short feedback question no longer holds it back.
• No time limit on the panel — it waits for you.
• If anything fails, your original prompt is sent unchanged.

Permissions are unchanged. An OpenAI API key is required.