Repository navigation
Pennant 0.4.0
Pennant 0.4.0
Faster, quieter and simpler: the chat answers in seconds, cards only come up for what can't be undone, and a new look.
- The chat answers in seconds. It used to wait seven seconds before even asking the model, while it looked through every recent message for cards waiting on you. That's instant now. Its prompt is laid out so the model can reuse most of it from one reply to the next, and starting, messaging or checking on a thread no longer takes seconds of its own.
- A model just for the chat. Settings › Advanced › Pennant › Model in the chat (or the model picker in the chat), or
pennant models chat <model>. Run the chat on a fast model and keep a stronger one for the work it hands to threads.pennant models threadsandpennant models fallbacksset the rest. - It says something first. When the chat hands work to a thread, its line ("On it, I'll let you know") shows straight away and the reply ends there.
- It gets ready while you type. A moment after you stop typing (or pause while talking to it on the iPhone), Pennant looks up what it knows about what you wrote and renews the model's sign-in, so sending goes straight to the model.
- It never goes silent. While work runs, the line above the message box and each thread's line in the chat say what it's doing in plain words ("Browsing the web", "Running a few commands", "Checking mail") and for how long, and when it's taking longer than usual.
- Cards only for what can't be undone. Settings › General › Ask me before… has three switches, sending and posting, spending money, and deletes that can't be undone, all off to start. Files Pennant deletes on your Mac go to the Trash. Every card has Dismiss, and a card nobody answers goes away after a day.
- Work doesn't stop at a card. A thread that needs your OK leaves the card holding exactly what it will do and carries on with everything else. Approve it and it runs as approved, and the thread hears how it went.
- A new look. The sidebar has the chat, your threads, Everything else (⌘1) and Settings. The chat is quieter: each reply's steps open from Show work, and each thread keeps one line that updates in place. Settings fit in four panes: General, Models, Connections and Advanced.
- Memory search runs on the Mac. Pennant searches what it remembers by meaning with its own embedding model, on the Mac's GPU, with nothing else to install. It embeds only what changed.
- Coding runs that change nothing say so. A run that only looked, or said what it means to do, ends as "Nothing changed" instead of "Done". After four runs in one piece of work, Pennant waits for your word before starting more.
- When a model is down, it moves on. A model that can't be reached hands over to the next one straight away instead of after minutes of retries, and a reply waiting on the model says that rather than that it failed.
- The iPhone app is on the App Store (not in the EU yet). Its next update shows your message the moment you send it, opens on what you were looking at while it connects, and catches up properly after being away. While you're at your Mac, your phone stays quiet; what Pennant has for you waits until you've been away a few minutes.
Requires macOS 15 or later, on Apple silicon or Intel. This is a preview; expect rough edges, and please report them.