Capy 0.4.3

Tasks are now called subagents, and on a Codex subscription the GPT-6 models use their full 1.05M-token context window.

This release also makes the desktop app start faster, and adds machine memory on the Billing page, a setting for where new messages scroll, the Firecrawl theme, and live turns in the API's thread stream. It fixes Setup, push notifications in Safari, and Test connection.

Subagents

A thread's child agents are now called subagents in the app, the CLI, Slack, and the docs. Prompts and automations that say task keep working, and the API still calls them tasks.

Full GPT-6 context on Codex

On a Codex subscription, GPT-6 Astra, GPT-6 Sol, GPT-6 Luna, and GPT-6.1 Sol now use their full 1.05M-token context window, the same window as on the OpenAI API. Before, a Codex subscription capped them at 400K. Fast mode stays at 272K.

Faster startup

The desktop app now opens its window before its background services finish loading.

Machine memory on Billing

The Billing page now has a Machine memory row in the Workspace card. It shows how many GB your running cloud machines use out of your plan's budget.

Where new messages scroll

The Appearance page has a new Chat section with Scroll new messages to top. Turn it off and the chat stays at the bottom when you send a message, instead of moving your message to the top.

Firecrawl theme

Firecrawl is a new built-in theme on the Appearance page, in light and dark.

Live turns in the API stream

A thread's stream in the public API now carries the turn in progress, so a client sees the reply as it's written and each tool as it's called.

Other fixes

A repository that pins its Node version now installs it from a prebuilt download before Setup runs, instead of compiling it from source. When Setup fails, Capy is told it failed instead of that it's still running. Push notifications now turn on in Safari and the iPhone Home Screen app, and Test connection shows the model and the provider's message when a connection is refused.