Capy 0.4.6
Private threads and projects
Threads and projects can now be private: share one with the people you choose, and nobody else in your organization sees it. You can also choose when each thread compacts, long sessions on GPT models run faster, and the thread chat is easier to read and steer.
This release also adds lazy repositories, share links for drive files, more than one subscription per provider, and Claude Haiku 5.5.
Private threads and projects
Set Access to Only specific people on a thread, a Captain, or a project, then add the people who should see it. Everyone you add gets a notification, and Transfer ownership hands a thread to someone else. On the Projects & setup page, Set owner picks who owns a project.
When you start a thread, the composer's Access menu makes it private and names its people before you send. The project picker lists your own projects first.
A teammate who follows a link to something private sees Ask for access instead of a missing page, and the owner decides. Admins choose who can keep work private, and the default access for new threads, in the organization's Sharing settings.
Compact at, per thread
Open a thread's context meter and click Compact at to choose when that thread compacts, anywhere up to the model's full window. Set org default makes the same value the default for your organization.
By default, a thread compacts at 400,000 tokens, or at 80% of a smaller model's window, and new threads use medium reasoning effort unless you pick another.
More than one subscription
Connect a second ChatGPT or SuperGrok subscription with Add another on the Models page. Your threads use your subscriptions in the order listed. When one runs out, a thread moves to the next one instead of leaving your own connections. Set a Usage cap on a subscription, and the next one takes over once it reaches that share of any limit.
Faster, steadier GPT sessions
Threads on GPT-6 models with a ChatGPT subscription start each step sooner, which adds up over a long session. After compaction, a thread keeps everything that was said and carries your open requests forward, so it picks up where it left off instead of stalling.
A new thread chat
An agent's reply now reads as sections. Hover a section and click Reply to quote it above the composer, so your answer is about that one point. The work between two messages folds into one line, and anything that needs you comes first on it.
A message you send from the app while a thread is working now steers it: the current step finishes, and Capy reads your message next instead of stopping. A queued message waits above the composer, where you can edit it, discard it, or send it now. Tool programs stay folded while they run, and interrupting a thread keeps the reasoning it already finished.
Other fixes
Mermaid labels no longer clip when zoomed, and a message sent to a waiting thread shows in the chat at once. Dictation keeps streaming after a send. A copied file link opens the pull request at that file, and a pull request tab never opens a pull request from another repository. Drive keeps your open file and scroll position, and Claude Opus 5.5 and Sonnet 5.5 threads on your own Anthropic key no longer stop with a signature error.