Agent Island 1.7.1 improves local Claude Code monitoring

Agent Island has released version 1.7.1 of its desktop companion for Claude Code and Codex, adding changes to local session monitoring on macOS and Windows.

The software shows developers the state of local AI coding sessions without sending session data to a hosted monitoring service. It reads supported local Claude Code and Codex session data and presents status information in a desktop interface.

That addresses a common issue for developers who leave coding agents running while they read documentation, review changes, or move between terminal windows and editors. A session may be working, waiting for input, finished, stalled, or paused by a rate limit, and those shifts may not be obvious when they happen.

Visual controls

Version 1.7.1 also expands the application's visual controls. Users can now choose teal, cobalt, violet, and silver glow colours, alongside Vivid and Calm visual modes.

The release adds a styled macOS installer and an in-app update prompt to make newer versions easier to find. It also includes reliability improvements in session monitoring and across the macOS interface.

Monitoring fixes

Among the changes, the app now avoids replayed Codex token events that could overstate totals. Archived Codex sessions are also included in local usage scans, improving the completeness of the usage view.

Alert behaviour has also been adjusted. Turn alerts now behave differently when the related terminal or editor is already in the foreground.

Additional macOS fixes cover panel behaviour, mouse-wheel handling, login-item reliability, and cost scanning. The focus of the release is to make the status display more dependable in day-to-day use rather than add a remote monitoring layer.

Local privacy

According to the project team, the application does not require an Agent Island account, has no product telemetry, and does not upload session data to Agent Island.

The project is distributed under the MIT Licence, and its source code is publicly available on GitHub. Agent Island is positioned as a companion to Claude Code and Codex rather than an alternative to either coding tool.

Its role is to answer a narrower operational question for developers running one or more coding sessions in the background: what is happening now, and does the developer need to return to that session? That distinction matters when different states, such as a finished task, an unanswered prompt, a stalled process, or a rate-limit pause, call for different responses.

Trusted status

Tristan Tang of the Agent Island team said the product is intended to keep that monitoring boundary on the user's own machine.

"AI coding sessions often finish or need input while the developer is focused somewhere else," said Tristan Tang, Agent Island Team, Agent Island. "Agent Island is designed to make that state visible at a glance without requiring developers to send their session data to another hosted monitoring service."

Tang said the latest work focused on whether the interface reflected local session activity accurately enough to remain useful over longer stretches of coding work.

"A status indicator is useful only when developers can trust that it reflects the underlying local session," said Tang. "The 1.7.1 work concentrated on details such as replayed events, archived sessions, foreground alerts, and usage scans because those details determine whether the display is genuinely useful during a long coding session."

He added that the project's aim is to keep the tool lightweight and avoid introducing another destination for development data.

"Developers should be able to understand the state of their local coding-agent work without creating another account or introducing a new data destination," added Tang. "Our goal is a small, transparent companion that helps people return to the right session at the right time."