Orca can keep idle chat agents running, and many terminals no longer crash it

Orca stopped every idle chat agent after thirty minutes, even with background work running. Now you choose fifteen minutes to a day, or Never. Pi chats also stay awake while subagents run.

Slide text

Orca now lets you choose when idle chat agents stop, or keep them running

Before

every chat agent stopped after 30 idle minutes

Now

15 minutes to 24 hours, or Never

Pi chats stay awake while their background subagents run.

Recent builds parsed the whole font file again for every terminal, so sessions with many terminals ran out of memory at launch. Now Orca uses its built-in ligature list, and thirty-two terminals relaunch without a crash.

Slide text

A session with many terminals no longer runs Orca out of memory at launch

Cause

ligature drawing parsed the whole font file again for every terminal

Now

Orca uses its built-in list of common ligatures

Test

32 terminals relaunched at 220 MB, with no crash

Font-specific ligatures outside that list may not join with the GPU renderer.

Seeing what Claude Code or Codex edited used to mean opening a tool group, then an edit card. Now each turn lists its changed files with line counts, and one click opens the patch.

Slide text

Each chat turn now lists the files your agent changed

Shows

files and line counts under the reply

Click

a filename opens its patch inline

Covers

Claude and Codex edits, on desktop and phone

When an agent answered slowly, Orca put your message back in the composer, and sending again delivered it twice. Now it goes out once, and long tool-heavy turns no longer stop halfway.

Slide text

A slow reply no longer sends your message twice or cuts a long turn short

Before

an unconfirmed message went back to the composer, and Send delivered it twice

Now

it stays in place, sent once, and a late reply attaches to it

Long turns

past 128 open text or tool entries, a healthy agent keeps going

Orca's browser now offers to save a login once it works, and fills it on your next visit. Without your system's secure storage the vault stays off, and paired-server pages don't get one.

Slide text

Orca’s browser now saves and fills your passwords

Import

from Chrome, Brave, Edge, Arc or a CSV file

Stored

encrypted with your system’s secure storage

No vault without secure storage, or on streamed paired-server pages and the web client.

Change Claude Code's default model, and new Orca chats now run it, instead of an old saved one. Managed Claude accounts keep your model and effort picks, and Claude chat gains Ultracode.

Slide text

New chats run the model your agent is configured for

Before

Orca sent an old saved default, even after you changed your config

Claude

managed accounts keep your /model and /effort picks

Also new

Ultracode in Claude chat’s effort menu

Until a chat starts, the picker can show the old default. Picks an older build overwrote are not recovered.

The experimental Orchestration panel is off again, even if you turned it on, because it could exhaust Orca's memory. Orchestration commands between agents keep working.

Slide text

The Orchestration panel is switched off again, for everyone

Why

it could exhaust Orca’s memory while agents worked unattended

Still works

chats, terminals and orchestration commands

Temporary, until the memory leak is repaired. Your saved opt-in is kept.

New installs pick Chat UI or Terminal, Orca Mobile's Home gets four tabs in the updated phone app, and switching worktrees is faster. If you keep lots of terminals open, I'd update today.

Slide text

Setup, Orca Mobile and worktrees get smaller changes

Setup

new installs choose Chat UI or Terminal

Orca Mobile

Home splits into Feed, Usage, Automations and Settings

Worktrees

switching between them is faster

Many terminals open? Update today.

Orca release notes on GitHub